Merge branch 'compression-core' of github.com:Egorich-print/OmniRoute into fix/combine-9115

The agentrouter.org upstream WAF returns 400 content-blocked
intermittently when:
  1. messages[].content contains a blocked keyword (Lorem ipsum, the
     phrase 'language model' alone, 'virtual assistant', etc.); or
  2. Requests from the same IP/key arrive in a burst, after which the
     WAF's per-IP suspicion bucket starts blocking content that would
     normally pass. The bucket relaxes after ~5-10s of idle.

Apply three mitigations:

1. Burst guard (open-sse/services/wafRateLimit.ts)
   Per-bucket (provider+url) gate that enforces a 500ms minimum gap
   between outbound requests to agentrouter. Configurable via
   configureWafRateLimit(). Tested in tests/unit/wafRateLimit.test.ts.

2. Reactive retry (BaseExecutor.WAF_RETRY_CONFIG in base.ts)
   New WAF_RETRY_CONFIG with maxAttempts=2, delayMs=1500,
   backoffMultiplier=2. When the upstream returns 400 with a body that
   matches /content[_-]blocked/i, retry the same URL with exponential
   backoff (1.5s, 3.0s) before falling through to the 429/401/fallback
   chain. Tested in tests/unit/base-executor-waf-retry.test.ts.

3. Documentation (docs/security/AGENTROUTER_WAF.md)
   Blocklist of always-blocked and almost-always-blocked patterns,
   behavior under load, guidance for prompts/tool output, and pointers
   to the relevant code paths in OmniRoute.

These are belt-and-suspenders: the burst guard prevents the WAF from
activating on normal traffic, and the reactive retry recovers when it
does anyway. Together they should eliminate the intermittent
400 content-blocked that Claude Code sees when running through
agentrouter via OmniRoute.

Refs #9275 follow-up. Test: 'WAF retry config shape' and 'WAF retry
differs from generic' guard the WAF_RETRY_CONFIG contract so future
refactors don't accidentally collapse the two retry paths.

Co-authored-by: diegosouzapw <diegosouzapw@users.noreply.github.com>

.cbmignore

@@ -0,0 +1,197 @@
+# codebase-memory-mcp ignore list
+#
+# Padrão gitignore-style. Linhas começando com `#` são comentários.
+# Barra final (`/`) = só diretório. Sem barra = casa arquivo OU diretório.
+#
+# O CBM também lê `.gitignore` automaticamente — esta lista deixa explícito o que
+# os hooks do CBM vão pular. Se uma regra entrar em conflito entre os dois arquivos,
+# vale a união. Editar este arquivo é mais barato do que confiar na herança implícita.
+#
+# Última reconciliação: 2026-07-31, status `ready` (513k nodes / 689k edges),
+# `auto_index_limit=50000`, total indexável medido ≈11.546 arquivos (folga 4,3×).
+#
+# Fontes cruzadas:
+#   - `codebase-memory-mcp cli index_status --project home-diegosouzapw-dev-proxys-OmniRoute`
+#     → `not_indexed.dirs` (27) + `not_indexed.files` (336), todos `BY DESIGN`.
+#   - `.gitignore` deste repo (5.691 B) — fonte canônica secundária.
+#
+# Como auditar mudanças: depois de editar este arquivo, rodar `index_repository`
+# (ou esperar `auto_watch` re-indexar) e re-checar `cli index_status` → comparar
+# contagens em `not_indexed.dirs_count` e `not_indexed.files_count`.
+
+# ─────────────────────────────────────────────────────────────────────────────
+# 1. Diretorios de runtime / pacote — nao sao codigo-fonte
+# ─────────────────────────────────────────────────────────────────────────────
+node_modules/
+node_modules
+
+# Builds e artefatos reproduziveis (Layer 1 Next.js / Electron)
+.build/
+dist/
+.next/
+out/
+
+# Electron especifico
+electron/dist-electron/
+electron/node_modules/
+icon.iconset/
+
+# Workspaces internos que tem proprio node_modules
+@omniroute/opencode-plugin/dist/
+@omniroute/opencode-plugin/node_modules/
+@omniroute/opencode-provider/dist/
+@omniroute/opencode-provider/node_modules/
+
+# Recursos nativos compilados (C/JNI/wasm)
+src/mitm/tproxy/native/build/
+
+# Artefatos locais do Stryker / Playwright / coverage
+.stryker-tmp/
+reports/mutation/
+stryker-output-*.json
+.playwright-mcp/
+test-results/
+playwright-report/
+blob-report/
+
+# Analise / linters / caches
+.analysis/
+.sisyphus/
+.plans/
+.gitnexus
+.worktrees
+.codegraph/
+
+# Quality artifacts (gerados por npm run lint --cache etc)
+.eslintcache
+.eslintcache-complexity
+
+# Claude Code local state
+.claude/scheduled_tasks.lock
+.claude/scheduled_tasks/
+.claude/sessions/
+.claude/state.json
+.claude/settings.local.json
+
+# Serena / Antigravity / outras tools locais
+.serena/
+.antigravitycli/
+.gemini/
+.config/
+
+# ─────────────────────────────────────────────────────────────────────────────
+# 2. Diretorios com prefixo `_` — locais / privados (regra global do .gitignore)
+# ─────────────────────────────────────────────────────────────────────────────
+_*/
+_artifacts/
+_cache/
+_mono_repo/
+_references/
+_tasks/
+
+# ─────────────────────────────────────────────────────────────────────────────
+# 3. Diretorios de tooling IA (state local, nao codigo)
+# ─────────────────────────────────────────────────────────────────────────────
+.agents/
+.claude/
+.vscode/
+.idea/
+.junie/
+.omc/
+.data/
+.data-dev/
+.local-data/
+.logs/
+.artifacts/
+.source/
+.superpowers/
+.claude-flow/
+.omnivscodeagent/
+omnirouteCloud/
+omnirouteSite/
+.omniroute/
+.stent/
+
+# Subpaths especificos do Claude Code que nao estao em .claude/ (criados sob repo)
+.claude/worktrees/
+
+# ─────────────────────────────────────────────────────────────────────────────
+# 4. Diretorios de dados / runtime locais (storage, env, secrets, scratch)
+# ─────────────────────────────────────────────────────────────────────────────
+data/
+src/lib/env/
+src/app/api/agent-skills/coverage/
+src/app/api/cloud/
+src/app/api/sync/cloud/
+src/app/api/system/env/
+tests/golden-set/data/
+
+# Logs e saida de teste
+logs/*
+test_output.log
+home-diegosouzapw-dev-automacoes-*.txt
+
+# ─────────────────────────────────────────────────────────────────────────────
+# 5. Diretorios do monorepo por subprojeto (nao fazem parte do app principal)
+# ─────────────────────────────────────────────────────────────────────────────
+security-analysis/
+vscode-extension/
+obsidian-plugin/node_modules/
+
+# ─────────────────────────────────────────────────────────────────────────────
+# 6. Diretorios de documentacao interna / workflow
+# ─────────────────────────────────────────────────────────────────────────────
+docs/superpowers/
+
+# ─────────────────────────────────────────────────────────────────────────────
+# 7. Arquivos especificos (nao diretorios inteiros)
+# ─────────────────────────────────────────────────────────────────────────────
+
+# Segredos e env — NUNCA indexar
+.env
+.env.*
+!.env.example
+!.env.homolog.example
+
+# TypeScript build info e next env declaration
+*.tsbuildinfo
+next-env.d.ts
+typescript
+
+# SQLite transient files (WAL/SHM/journal)
+*.sqlite-shm
+*.sqlite-wal
+*.sqlite-journal
+
+# Mapas e source maps
+*.map
+
+# Bun / npm lockfiles ruidosos
+bun.lock
+
+# `cheaper-inference-gateway.svg` e arquivos de midia na raiz/asset ja cobertos
+# pelos `ignored-suffix` do indexador (svg/png/jpg/ico/etc >50kB ou >500linhas);
+# manter a regra explicita aqui ajuda a auditar:
+cheaper-inference-gateway.svg
+cheaper-inference-gateway-*.svg
+
+# Husky internals
+.husky/_/
+
+# CI / quality metric artifacts
+config/quality/quality-metrics.json
+config/quality/test-impact-map.json
+audit-report.json
+.gh-discussions.json
+
+# i18n audit (gerado por npm run scripts)
+scripts/i18n/_audit.json
+scripts/i18n/_pending-keys.json
+
+# Cli binario local (scratch)
+bin/omniroute.mjs
+
+# Deploy / docker backups
+deploy.sh
+docker-compose.yml.bak
+docker-compose.minimal.yml

.dockerignore

@@ -18,6 +18,7 @@ coverage
 # Runtime data and logs
 data
 logs
+.sandbox
 
 # Local env files (inject at runtime via --env-file or -e)
 .env

.env.devin-bridge.example

@@ -0,0 +1,6 @@
+ENABLE_LIVE_DEVIN_TESTS=0
+DEVIN_BRIDGE_MODEL=devin-cli-agentic/swe-1-7
+DEVIN_BRIDGE_SONNET_MODEL=devin-cli-agentic/swe-1-7
+DEVIN_BRIDGE_OPUS_MODEL=devin-cli-agentic/swe-1-7
+DEVIN_BRIDGE_HAIKU_MODEL=devin-cli-agentic/swe-1-7
+DEVIN_BRIDGE_SUBAGENT_MODEL=devin-cli-agentic/swe-1-7

.env.example

@@ -1414,10 +1414,6 @@ APP_LOG_TO_FILE=true
 # Default: ~/.omniroute/plugins/  Override in dev/CI to point at a local plugin tree.
 # OMNIROUTE_PLUGIN_PATH=
 
-# Allow plugins to request the 'exec' permission (spawn child processes from the
-# plugin worker sandbox). Disabled by default; set to 1 to enable (local operator only).
-# OMNIROUTE_PLUGINS_ALLOW_EXEC=0
-
 # ── Prompt cache (system prompt deduplication) ──
 # Used by: open-sse/services — caches identical system prompts across requests.
 # PROMPT_CACHE_MAX_SIZE=50              # Max cached entries (default: 50)
@@ -1843,6 +1839,18 @@ APP_LOG_TO_FILE=true
 # ── Devin CLI binary path ──
 # Used by: open-sse/executors/devin-cli.ts. Default: looked up via PATH.
 # CLI_DEVIN_BIN=devin
+# Agentic bridge-only binary override. The bridge still executes ACP stdio only.
+# CLI_DEVIN_AGENTIC_BIN=devin
+# Required isolated HOME for the agentic Devin child process.
+# DEVIN_AGENTIC_HOME=/home/bridge
+# Bounded ACP turn timeout in milliseconds. Default: 120000.
+# DEVIN_AGENTIC_ACP_TIMEOUT_MS=120000
+# Agentic bridge model aliases. Values must keep the devin-cli-agentic/ prefix.
+# DEVIN_BRIDGE_MODEL=devin-cli-agentic/swe-1-7
+# DEVIN_BRIDGE_SONNET_MODEL=devin-cli-agentic/swe-1-7
+# DEVIN_BRIDGE_OPUS_MODEL=devin-cli-agentic/swe-1-7
+# DEVIN_BRIDGE_HAIKU_MODEL=devin-cli-agentic/swe-1-7
+# DEVIN_BRIDGE_SUBAGENT_MODEL=devin-cli-agentic/swe-1-7
 
 # ── Command Code (custom CLI) callback ──
 # Local port used for OAuth-style callbacks from the Command Code CLI helper.
@@ -2304,6 +2312,18 @@ QUOTA_STORE_DRIVER=sqlite              # sqlite | redis
 # ─────────────────────────────────────────────────────────────────────────────
 # HYPERAGENT_USAGE_URL=https://hyperagent.com/api/settings/billing/usage
 
+# ─────────────────────────────────────────────────────────────────────────────
+# ChatGPT Web (Codex) headless browser and outbound tool tunnel
+# Used by: open-sse/executors/chatgpt-web-codex.ts
+# Connection values entered in the dashboard override these global defaults.
+# ─────────────────────────────────────────────────────────────────────────────
+# CHATGPT_WEB_CODEX_CHROME_PATH=/usr/bin/chromium
+# CHROME_PATH=/usr/bin/chromium
+# CHATGPT_WEB_CODEX_CDP_URL=http://chatgpt-web-codex-browser:9223
+# CHATGPT_WEB_CODEX_TUNNEL_ID=tunnel_0123456789abcdef0123456789abcdef
+# CHATGPT_WEB_CODEX_RUNTIME_KEY=
+# CHATGPT_WEB_CODEX_CONNECTOR_NAME=OmniRoute Codex
+
 # ─────────────────────────────────────────────────────────────────────────────
 # Browser-login VNC sessions (optional — src/lib/vncSession/manifest.ts)
 # Containerized Chromium+VNC used for interactive browser-login credential

.github/pull_request_template.md

@@ -9,11 +9,14 @@
 
 ## Validation
 
-Run only the focused loop for what you changed — the full unit suite, Vitest, the
+Choose the change type and focused loop from the
-60% coverage gate, and the production build all run in CI on this PR (#8329):
+[Contribution Golden Path](../docs/dev/CONTRIBUTION_GOLDEN_PATH.md). The full unit suite,
+Vitest, the 60% coverage gate, and the production build all run in CI on this PR (#8329):
 
-- [ ] Focused tests for the change: `node --import tsx/esm --test tests/unit/<file>.test.ts`
+- [ ] Change type: provider / routing / UI / i18n / CLI / DB / build-deploy / other
+- [ ] Focused tests and category gates from the golden path
 - [ ] `npm run lint`
+- [ ] Reconciled with the current active release base; focused checks rerun afterward
 - [ ] Production-code changes include a new or updated automated test in this PR
 - [ ] SonarQube PR analysis is green or any remaining issues are explicitly documented below
 
@@ -29,4 +32,4 @@ Run only the focused loop for what you changed — the full unit suite, Vitest,
 
 ## Reviewer Notes
 
-- Call out any risky areas, migrations, feature flags, or manual validation that reviewers should know about.
+- Call out any risky areas, migrations, feature flags, or manual validation that reviewers should know about.

.github/workflows/ci.yml

@@ -27,7 +27,7 @@ env:
 jobs:
   changes:
     name: Change Classification
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     outputs:
       code: ${{ steps.classify.outputs.code }}
       docs: ${{ steps.classify.outputs.docs }}
@@ -35,13 +35,10 @@ jobs:
       workflow: ${{ steps.classify.outputs.workflow }}
       testsOnly: ${{ steps.classify.outputs.testsOnly }}
     steps:
-      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           fetch-depth: 0
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
       # Refuse a PR that targets its own head branch before spending anything on it. #8912 has
       # head == base == release/v3.8.50: no diff, can never merge, and it sits in the queue with
       # a full check board attached on every push to that branch. One field comparison.
@@ -77,7 +74,7 @@ jobs:
 
   lint:
     name: Lint
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     needs: changes
     # P3 (plano mestre): a release-PR viva fica DRAFT o ciclo inteiro — jobs pesados pulam
     # drafts (ciclo v3.8.44: 123 runs pesados re-disparados por merges na release, 88 cancelados).
@@ -91,13 +88,9 @@ jobs:
       API_KEY_SECRET: ci-lint-api-key-secret-long
       DISABLE_SQLITE_AUTO_BACKUP: "true"
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       - run: npm run check:node-runtime
       - run: npm run audit:deps
@@ -171,7 +164,7 @@ jobs:
 
   quality-gate:
     name: Quality Ratchet
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     # needs lint so eslint-results artifact is available (same inventory as the
     # blocking lint step). Allow lint failure so other ratchets still run.
     needs: [changes, test-coverage, lint]
@@ -191,13 +184,9 @@ jobs:
       contents: read
       security-events: read
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       - name: Restore ESLint file cache
         uses: actions/cache@v6
@@ -289,7 +278,7 @@ jobs:
   # SonarQube needs SONAR_TOKEN/SONAR_HOST_URL secrets.
   quality-extended:
     name: Quality Gates (Extended)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     needs: changes
     # P3 (plano mestre): a release-PR viva fica DRAFT o ciclo inteiro — jobs pesados pulam
     # drafts (ciclo v3.8.44: 123 runs pesados re-disparados por merges na release, 88 cancelados).
@@ -299,14 +288,10 @@ jobs:
       # fetch-depth: 0 — the OpenAPI breaking-change gate (oasdiff) reads the base
       # spec via `git show <base_ref>:docs/openapi.yaml`; a shallow clone
       # would lack the base ref and the gate would self-skip (base-unresolved).
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           fetch-depth: 0
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       # Dead-code, cognitive-complexity, type-coverage foram promovidos ao job
       # quality-gate (bloqueante) na Fase 7 INT — não rodam aqui para evitar duplo custo.
@@ -409,20 +394,16 @@ jobs:
 
   docs-sync-strict:
     name: Docs Sync (Strict)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     needs: changes
     # P3 (plano mestre): a release-PR viva fica DRAFT o ciclo inteiro — jobs pesados pulam
     # drafts (ciclo v3.8.44: 123 runs pesados re-disparados por merges na release, 88 cancelados).
     # Run when docs OR code change: API/route code can break doc/OpenAPI contract gates.
     if: ${{ github.event_name != 'pull_request' || (github.event.pull_request.draft == false && (needs.changes.outputs.docs == 'true' || needs.changes.outputs.code == 'true')) }}
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       - run: npm run check:docs-all
       # Previously-orphaned contract gates (existed as files, never wired anywhere).
@@ -442,7 +423,7 @@ jobs:
 
   docs-lint:
     name: Docs Lint (prose — advisory)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     needs: changes
     # P3 (plano mestre): a release-PR viva fica DRAFT o ciclo inteiro — jobs pesados pulam
     # drafts (ciclo v3.8.44: 123 runs pesados re-disparados por merges na release, 88 cancelados).
@@ -452,13 +433,9 @@ jobs:
     # existing doc corpus is brought up to style. Promote to blocking once it converges.
     continue-on-error: true
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - name: markdownlint (docs + root, advisory)
         run: npx --yes markdownlint-cli2 "docs/**/*.md" "*.md" "!docs/i18n" "!docs/research" || true
       - name: Vale prose lint (Microsoft style, advisory)
@@ -473,7 +450,7 @@ jobs:
 
   i18n-ui-coverage:
     name: i18n UI Coverage
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     needs: changes
     # P3 (plano mestre): a release-PR viva fica DRAFT o ciclo inteiro — jobs pesados pulam
     # drafts (ciclo v3.8.44: 123 runs pesados re-disparados por merges na release, 88 cancelados).
@@ -483,14 +460,10 @@ jobs:
       # fetch-depth: 0 — the value-drift gate diffs en.json against the merge base to
       # find rewritten English strings. On a shallow clone the base ref is missing and
       # the gate self-skips (base-unresolved), so it would never actually run.
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           fetch-depth: 0
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       - run: node scripts/i18n/check-ui-keys-coverage.mjs --threshold=65
       # #8463: a rewritten English value used to leave its 39 translations behind
@@ -506,17 +479,13 @@ jobs:
   # without needing app-boot/Playwright infra. Same gating as i18n-ui-coverage.
   i18n-glossary-zhcn:
     name: i18n Glossary (zh-CN)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     needs: changes
     if: ${{ github.event_name != 'pull_request' || (github.event.pull_request.draft == false && (needs.changes.outputs.i18n == 'true' || needs.changes.outputs.code == 'true')) }}
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       - run: node scripts/i18n/check-glossary-consistency.mjs --locale=zh-CN
       - run: node scripts/i18n/check-glossary-consistency.mjs --locale=zh-TW
@@ -528,7 +497,7 @@ jobs:
   # idioma (a matrix antiga subia 40 artifacts cujo result.txt colidia no merge-multiple).
   i18n:
     name: i18n Validation (all languages)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     needs: changes
     # P3 (plano mestre): a release-PR viva fica DRAFT o ciclo inteiro — jobs pesados pulam
     # drafts (ciclo v3.8.44: 123 runs pesados re-disparados por merges na release, 88 cancelados).
@@ -536,7 +505,7 @@ jobs:
     if: ${{ github.event_name != 'pull_request' || (github.event.pull_request.draft == false && needs.changes.outputs.i18n == 'true') }}
     continue-on-error: true
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
       - uses: actions/setup-python@v7
@@ -571,15 +540,12 @@ jobs:
   pr-test-policy:
     name: PR Test Policy
     if: ${{ github.event_name == 'pull_request' && github.event.pull_request.draft == false }}
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           fetch-depth: 0
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
       - name: Fetch base branch
         run: git fetch --no-tags origin "${GITHUB_BASE_REF}"
       - name: Validate source changes include tests
@@ -606,17 +572,17 @@ jobs:
     # online), the heavy jobs run on the dedicated 32-core VPS runners (label
     # omni-release) instead of queueing on the 20-concurrent-job hosted pool.
     # Safety: fork PRs NEVER reach the self-hosted runner — the expression falls
-    # back to ubuntu-latest unless the PR head repo is this repository (push /
+    # back to ubuntu-26.04 unless the PR head repo is this repository (push /
     # dispatch events are own-origin by definition). Any failure path (VM down,
-    # var unset/false) also falls back to ubuntu-latest.
+    # var unset/false) also falls back to ubuntu-26.04.
-    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository)) && fromJSON('["self-hosted","omni-release"]') || 'ubuntu-latest' }}
+    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository)) && fromJSON('["self-hosted","omni-release"]') || 'ubuntu-26.04' }}
     needs: changes
     if: ${{ github.event_name != 'pull_request' || (needs.changes.outputs.code == 'true' && github.event.pull_request.draft == false) }}
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
         with:
           node-version: ${{ env.CI_NODE_VERSION }}
           cache: npm
@@ -656,18 +622,14 @@ jobs:
 
   package-artifact:
     name: Package Artifact
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     needs: build
     env:
       JWT_SECRET: ci-build-secret-with-sufficient-length-for-validation
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       - run: npm run check:node-runtime
       - name: Download Next.js build artifact
@@ -703,15 +665,15 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-latest, windows-latest]
+        os: [ubuntu-26.04, windows-latest]
     env:
       JWT_SECRET: ci-build-secret-with-sufficient-length-for-validation
       CSC_IDENTITY_AUTO_DISCOVERY: "false"
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
         with:
           node-version: ${{ env.CI_NODE_VERSION }}
           cache: npm
@@ -750,14 +712,14 @@ jobs:
 
   test-unit:
     name: Unit Tests (${{ matrix.shard }}/8)
-    # Same dynamic-runner rule as Build (own-origin only; fallback ubuntu-latest).
+    # Same dynamic-runner rule as Build (own-origin only; fallback ubuntu-26.04).
     # PINNED to hosted, deliberately not on the USE_VPS_RUNNER switch (gap 19). One variable
     # governed the build and the test jobs, which want OPPOSITE machines: the build needs the
     # .113's RAM, the tests need the hosted runner's link. Measured on 2026-07-29 —
     # actions/setup-node took 20m06s on .113 with 4 concurrent runners versus 16s hosted (npm
     # cache restore saturating the link), while the tests themselves tied, 2m54 vs 2m31. So
     # self-hosted is strictly worse here and there is nothing to configure.
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     timeout-minutes: 25
     # needs: changes (not build) — this job never downloads the next-build artifact;
     # gating it on Build only serialized ~20min of wall-clock for nothing. Jobs that
@@ -775,13 +737,9 @@ jobs:
       API_KEY_SECRET: ci-test-api-key-secret-long
       DISABLE_SQLITE_AUTO_BACKUP: "true"
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       - run: npm run check:node-runtime
       # QW-d (plano mestre): fonte única — o MESMO npm script dos runs locais (adiciona o
@@ -811,31 +769,27 @@ jobs:
 
   test-bun-sqlite:
     name: Bun SQLite Compatibility
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     timeout-minutes: 10
     needs: changes
     if: ${{ github.event_name != 'pull_request' || (needs.changes.outputs.code == 'true' && github.event.pull_request.draft == false) }}
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       - run: npm run test:bun:db
 
   test-vitest:
     name: Vitest (MCP / autoCombo / UI components)
-    # Same dynamic-runner rule as Build (own-origin only; fallback ubuntu-latest).
+    # Same dynamic-runner rule as Build (own-origin only; fallback ubuntu-26.04).
     # PINNED to hosted, deliberately not on the USE_VPS_RUNNER switch (gap 19). One variable
     # governed the build and the test jobs, which want OPPOSITE machines: the build needs the
     # .113's RAM, the tests need the hosted runner's link. Measured on 2026-07-29 —
     # actions/setup-node took 20m06s on .113 with 4 concurrent runners versus 16s hosted (npm
     # cache restore saturating the link), while the tests themselves tied, 2m54 vs 2m31. So
     # self-hosted is strictly worse here and there is nothing to configure.
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     timeout-minutes: 15
     # needs: changes (not build) — no artifact consumed; see test-unit note.
     needs: changes
@@ -845,13 +799,9 @@ jobs:
       API_KEY_SECRET: ci-test-api-key-secret-long
       DISABLE_SQLITE_AUTO_BACKUP: "true"
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       # The second test runner (CLAUDE.md: "Both test runners must pass") — was never
       # wired into CI until the 2026-06-09 quality audit (Fase 6A.2).
@@ -879,7 +829,7 @@ jobs:
   # the release gate can still exercise them via workflow_dispatch when needed).
   test-coverage:
     name: Coverage
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     # 10min was sized before #7114 added the lcov reporter (Codecov/Sonar need it);
     # merging 8 shard JSONs + text+json+lcov now takes ~10-12min — three consecutive
     # release-tip runs died at exactly 10m as job-timeout "cancelled" (2026-07-15/16).
@@ -890,13 +840,9 @@ jobs:
       JWT_SECRET: ci-test-secret-with-sufficient-length-for-validation
       API_KEY_SECRET: ci-test-api-key-secret-long
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       - name: Download all shard coverage
         uses: actions/download-artifact@v8
@@ -980,14 +926,14 @@ jobs:
 
   sonarqube:
     name: SonarQube
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     needs: test-coverage
     if: ${{ !cancelled() && needs.test-coverage.result == 'success' }}
     env:
       SONAR_TOKEN: ${{ secrets.SONAR_TOKEN }}
       SONAR_HOST_URL: ${{ secrets.SONAR_HOST_URL }}
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           fetch-depth: 0
@@ -1032,7 +978,7 @@ jobs:
 
   coverage-pr-comment:
     name: PR Coverage Comment
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     if: ${{ !cancelled() && github.event_name == 'pull_request' && github.event.pull_request.draft == false && github.event.pull_request.head.repo.fork == false && needs.changes.outputs.code == 'true' }}
     needs:
       - changes
@@ -1111,7 +1057,7 @@ jobs:
 
   test-e2e:
     name: E2E Tests (${{ matrix.shard }}/9)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     # Build artifact from the `build` job is downloaded instead of rebuilding
     # (~5min saved per shard). 9 shards (up from 6) reduces tests per shard by
     # ~33%. Playwright browser is cached across runs (~1.5min saved per shard).
@@ -1133,13 +1079,9 @@ jobs:
       DISABLE_SQLITE_AUTO_BACKUP: "true"
       OMNIROUTE_PLAYWRIGHT_SKIP_BUILD: "1"
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       - run: npm run check:node-runtime
       - name: Cache Playwright browsers
@@ -1188,7 +1130,7 @@ jobs:
 
   test-integration:
     name: Integration Tests (${{ matrix.shard }}/2)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     timeout-minutes: 15
     # needs: changes (not build) — no artifact consumed; see test-unit note.
     needs: changes
@@ -1204,13 +1146,9 @@ jobs:
       DATA_DIR: /tmp/omniroute-ci-${{ matrix.shard }}
       DISABLE_SQLITE_AUTO_BACKUP: "true"
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       - run: npm run check:node-runtime
       # (tsx/esm = QW-b; o alinhamento de ESCOPO do integration com o npm script fica p/ follow-up)
@@ -1218,7 +1156,7 @@ jobs:
 
   test-security:
     name: Security Tests
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     # needs: changes (not build) — no artifact consumed; see test-unit note.
     needs: changes
     if: ${{ github.event_name != 'pull_request' || (needs.changes.outputs.code == 'true' && github.event.pull_request.draft == false) }}
@@ -1227,20 +1165,16 @@ jobs:
       API_KEY_SECRET: ci-test-api-key-secret-long
       DISABLE_SQLITE_AUTO_BACKUP: "true"
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       - run: npm run check:node-runtime
       - run: npm run test:security
 
   ci-summary:
     name: CI Dashboard
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     if: ${{ !cancelled() }}
     needs:
       - changes

.github/workflows/claude.yml

@@ -21,7 +21,7 @@ jobs:
       (github.event_name == 'pull_request_review_comment' && contains(github.event.comment.body, '@claude')) ||
       (github.event_name == 'pull_request_review' && contains(github.event.review.body, '@claude')) ||
       (github.event_name == 'issues' && (contains(github.event.issue.body, '@claude') || contains(github.event.issue.title, '@claude')))
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     permissions:
       contents: read
       pull-requests: read
@@ -30,7 +30,7 @@ jobs:
       actions: read # Required for Claude to read CI results on PRs
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           fetch-depth: 1

.github/workflows/codeql.yml

@@ -13,7 +13,7 @@ permissions:
 jobs:
   analyze:
     name: Analyze (javascript-typescript)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     permissions:
       security-events: write
       actions: read

.github/workflows/dast-smoke.yml

@@ -18,7 +18,7 @@ concurrency:
   cancel-in-progress: true
 jobs:
   dast-smoke:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     # ADVISORY while this new gate matures (repo convention: advisory -> blocking).
     # Flip to blocking (remove continue-on-error) once it's proven stable across a few PRs.
     continue-on-error: true
@@ -33,10 +33,6 @@ jobs:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
-        with:
-          node-version: "24"
-          cache: npm
       - run: npm ci
       - name: Build CLI bundle
         env:

.github/workflows/deploy-vps.yml

@@ -15,7 +15,7 @@ jobs:
       (github.event_name == 'workflow_dispatch' || github.event.workflow_run.conclusion == 'success')
       && vars.DEPLOY_ENABLED == 'true'
     name: Deploy OmniRoute to VPS
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
       - name: Check VPS SSH reachability from runner
         id: reach

.github/workflows/docker-publish.yml

@@ -33,7 +33,7 @@ permissions:
 jobs:
   prepare:
     name: Resolve Docker release metadata
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     outputs:
       version: ${{ steps.version.outputs.version }}
       promote_latest: ${{ steps.version.outputs.promote_latest }}
@@ -42,7 +42,7 @@ jobs:
       IMAGE_NAME: diegosouzapw/omniroute
     steps:
       - name: Checkout
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           ref: ${{ github.event_name == 'workflow_dispatch' && format('refs/tags/v{0}', inputs.version) || '' }}
@@ -145,7 +145,7 @@ jobs:
       GHCR_IMAGE_NAME: ghcr.io/diegosouzapw/omniroute
     steps:
       - name: Checkout
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           ref: ${{ github.event_name == 'workflow_dispatch' && format('refs/tags/v{0}', inputs.version) || '' }}
@@ -233,7 +233,7 @@ jobs:
       - prepare
       - build
     if: needs.prepare.outputs.skip != 'true'
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     permissions:
       contents: read
       packages: write
@@ -245,7 +245,7 @@ jobs:
       PROMOTE_LATEST: ${{ needs.prepare.outputs.promote_latest }}
     steps:
       - name: Checkout
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           ref: ${{ github.event_name == 'workflow_dispatch' && format('refs/tags/v{0}', inputs.version) || '' }}

.github/workflows/electron-release.yml

@@ -20,14 +20,14 @@ permissions:
 jobs:
   validate:
     name: Validate version
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     permissions:
       contents: read
     outputs:
       version: ${{ steps.validate.outputs.version }}
     steps:
       - name: Checkout code
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           fetch-depth: 0
@@ -78,17 +78,17 @@ jobs:
             target: mac-arm64
             ext: -arm64.dmg
           - platform: linux
-            runner: ubuntu-latest
+            runner: ubuntu-26.04
             target: linux
             ext: .AppImage
             deb_ext: .deb
 
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
       - name: Setup Node
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
         with:
           node-version: 24
           cache: npm
@@ -239,12 +239,12 @@ jobs:
     # Now: attach everything that did build, then fail the job loudly (see the last
     # step) so an incomplete channel is visible instead of silent.
     if: ${{ !cancelled() && needs.validate.result == 'success' }}
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     permissions:
       contents: write # softprops/action-gh-release creates the GitHub Release
     steps:
       - name: Checkout
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           fetch-depth: 0
@@ -329,7 +329,7 @@ jobs:
     # of passing unnoticed — the v3.8.49 release had ZERO assets and every gate was
     # green, because nothing ever asserted the release HAS binaries.
     if: ${{ !cancelled() && needs.release.result == 'success' }}
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     permissions:
       contents: read
     steps:

.github/workflows/lock-released-branch.yml

@@ -40,7 +40,7 @@ jobs:
   # ─────────────────────────────────────────────────────────────────────────
   lock-branch:
     if: github.event_name == 'release' || github.event_name == 'workflow_dispatch'
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
       - name: Lock release/<tag> branch
         env:
@@ -97,7 +97,7 @@ jobs:
   # ─────────────────────────────────────────────────────────────────────────
   guard-no-push-after-release:
     if: github.event_name == 'push'
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
       - name: Reject push if matching release tag exists
         env:

.github/workflows/mutation-redundancy.yml

@@ -22,7 +22,7 @@ permissions:
 jobs:
   stryker-nobail:
     name: Stryker disableBail (batch ${{ matrix.batch.name }})
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     strategy:
       fail-fast: false
       matrix:
@@ -41,13 +41,9 @@ jobs:
             mutate: "open-sse/handlers/chatCore/telemetryHelpers.ts,open-sse/handlers/chatCore/memorySkillsInjection.ts,open-sse/handlers/chatCore/semanticCache.ts"
     timeout-minutes: 300
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: "24"
-          cache: npm
       - run: npm ci
       - name: Run Stryker (disableBail)
         env:

.github/workflows/nightly-compat.yml

@@ -28,11 +28,11 @@ concurrency:
 jobs:
   resolve-branch:
     name: Resolve active release branch
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     outputs:
       target: ${{ steps.branch.outputs.target }}
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           fetch-depth: 0
           persist-credentials: false
@@ -58,15 +58,15 @@ jobs:
 
   compat-build-26:
     name: Node 26 Compatibility Build
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     timeout-minutes: 25
     needs: resolve-branch
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           ref: ${{ needs.resolve-branch.outputs.target }}
           persist-credentials: false
-      - uses: actions/setup-node@v7
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
         with:
           node-version: "26"
           cache: npm
@@ -75,7 +75,7 @@ jobs:
       # CI_NODE_VERSION=24). It failed every nightly with the runner-reclaimed
       # signature ("The runner has received a shutdown signal" / "The operation was
       # canceled", no exit code) always at the same Turbopack compile phase — a
-      # classic OOM kill on the memory-constrained ubuntu-latest runner. Turbopack's
+      # classic OOM kill on the memory-constrained 16 GB hosted runner. Turbopack's
       # native (Rust, off-V8-heap) allocation is NOT bounded by --max-old-space-size
       # and peaks far higher than webpack on this large module graph (#6409), and is
       # heavier still under Node 26. Use the documented webpack fallback here: it still
@@ -88,7 +88,7 @@ jobs:
 
   compat-tests:
     name: Node ${{ matrix.node }} Compat Tests (${{ matrix.shard }}/4)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     timeout-minutes: 25
     needs: resolve-branch
     strategy:
@@ -102,11 +102,11 @@ jobs:
       DISABLE_SQLITE_AUTO_BACKUP: "true"
       TEST_SHARD: ${{ matrix.shard }}/4
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           ref: ${{ needs.resolve-branch.outputs.target }}
           persist-credentials: false
-      - uses: actions/setup-node@v7
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
         with:
           node-version: ${{ matrix.node }}
           cache: npm
@@ -116,7 +116,7 @@ jobs:
 
   report:
     name: Open / update tracking issue on failure
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     if: ${{ !cancelled() && (needs.compat-tests.result == 'failure' || needs.compat-build-26.result == 'failure') }}
     needs: [resolve-branch, compat-build-26, compat-tests]
     permissions:

.github/workflows/nightly-llm-security.yml

@@ -10,13 +10,11 @@ permissions:
 jobs:
   promptfoo-guard:
     name: promptfoo — injection guard (block mode, no secret)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with: { node-version: "24", cache: npm }
       - run: npm ci
       - name: Build CLI bundle
         env:
@@ -46,7 +44,7 @@ jobs:
 
   garak:
     name: garak probes (skip without provider secret)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     # NOTE: the `secrets` context is NOT available in a job-level `if:` — referencing
     # it there makes GitHub reject the file on push (startup_failure on every push).
     # Map the secret into a job-level env and gate each step on a presence check, so
@@ -63,13 +61,10 @@ jobs:
             echo "run=false" >> "$GITHUB_OUTPUT"
             echo "::notice::PROMPTFOO_PROVIDER_KEY not set — skipping garak probes (advisory)."
           fi
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
         if: steps.gate.outputs.run == 'true'
-      - uses: actions/setup-node@v7
-        if: steps.gate.outputs.run == 'true'
-        with: { node-version: "24", cache: npm }
       - run: npm ci
         if: steps.gate.outputs.run == 'true'
       - name: Build CLI bundle

.github/workflows/nightly-mutation.yml

@@ -10,7 +10,7 @@ permissions:
 jobs:
   stryker:
     name: Stryker mutation (batch ${{ matrix.batch.name }} — advisory)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     # Mutation testing is expensive. History of the budget:
     #   - Full 8-module set TIMED OUT at the 180min cap (run 27705123780 = exactly 180min).
     #     The two god-files chatCore.ts/combo.ts dominated ~2/3 of the mutants and were
@@ -104,13 +104,9 @@ jobs:
     # scripts/quality/mutation-radiography.mjs both merge per file).
     timeout-minutes: ${{ matrix.batch.timeout || 180 }}
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: "24"
-          cache: npm
       - run: npm ci
       - name: Restore Stryker incremental cache
         uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
@@ -145,15 +141,12 @@ jobs:
     name: Mutation score ratchet (blocking)
     needs: stryker
     if: always()
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     timeout-minutes: 15
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: "24"
       - name: Download all mutation reports
         uses: actions/download-artifact@v8
         with:

.github/workflows/nightly-property.yml

@@ -8,15 +8,11 @@ permissions:
   issues: write
 jobs:
   property-random-seed:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: "24"
-          cache: npm
       - run: npm ci
       - name: fast-check random seed (high runs)
         id: prop

.github/workflows/nightly-release-green.yml

@@ -68,13 +68,13 @@ jobs:
     # this runs on the dedicated VPS runner — clean env (no operator OMNIROUTE_API_KEY,
     # no local noauth CLIs => zero machine-specific false positives) and no contention.
     # Nightly cron normally finds the var false (VM off) and falls back to hosted.
-    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && fromJSON('["self-hosted","omni-release"]')) || 'ubuntu-latest' }}
+    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && fromJSON('["self-hosted","omni-release"]')) || 'ubuntu-26.04' }}
     env:
       JWT_SECRET: ci-nightly-secret-with-sufficient-length-for-validation
       API_KEY_SECRET: ci-nightly-api-key-secret-long
       DISABLE_SQLITE_AUTO_BACKUP: "true"
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           fetch-depth: 0
           persist-credentials: false
@@ -116,7 +116,7 @@ jobs:
           git checkout "$TARGET"
           git log -1 --oneline
 
-      - uses: actions/setup-node@v7
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
         with:
           node-version: "24"
           cache: npm
@@ -217,19 +217,19 @@ jobs:
     # On a push, only run for a push to main — a push to release/* is handled by
     # release-green above. Schedule/dispatch always run (they also sweep main).
     if: ${{ github.event_name != 'push' || github.ref_name == 'main' }}
-    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && fromJSON('["self-hosted","omni-release"]')) || 'ubuntu-latest' }}
+    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && fromJSON('["self-hosted","omni-release"]')) || 'ubuntu-26.04' }}
     env:
       JWT_SECRET: ci-nightly-secret-with-sufficient-length-for-validation
       API_KEY_SECRET: ci-nightly-api-key-secret-long
       DISABLE_SQLITE_AUTO_BACKUP: "true"
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           ref: main # literal — no injection surface; scheduled runs default to the repo default branch (a release/v*), so pin main explicitly
           fetch-depth: 0
           persist-credentials: false
 
-      - uses: actions/setup-node@v7
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
         with:
           node-version: "24"
           cache: npm
@@ -331,12 +331,12 @@ jobs:
   bank-ratchet-shrinks:
     name: Bank ratchet shrinks
     if: ${{ github.event_name != 'push' }}
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     permissions:
       contents: write
       pull-requests: write
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           fetch-depth: 0
 
@@ -371,11 +371,6 @@ jobs:
           git checkout "$TARGET"
           git log -1 --oneline
 
-      - uses: actions/setup-node@v7
-        with:
-          node-version: "24"
-          cache: npm
-
       - uses: ./.github/actions/npm-ci-retry
 
       - name: Ratchet the baselines down

.github/workflows/nightly-resilience.yml

@@ -10,43 +10,31 @@ permissions:
 jobs:
   heap:
     name: Heap-growth gate
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: "24"
-          cache: npm
       - run: npm ci
       - run: npm run test:heap
 
   chaos:
     name: Resilience chaos (fault injection)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: "24"
-          cache: npm
       - run: npm ci
       - run: npm run test:chaos
 
   k6-soak:
     name: k6 load/soak
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: "24"
-          cache: npm
       - run: npm ci
       - name: Build CLI bundle
         env:
@@ -78,7 +66,7 @@ jobs:
 
   a11y:
     name: A11y axe (nightly, freeze-and-alert)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     # The Playwright webServer (`start` mode) builds Next via build-next-isolated.mjs and
     # boots the standalone server itself (waits on /api/monitoring/health, 15min webServer
     # timeout). Unlike the per-PR test-e2e job, this nightly job has no pre-built artifact,
@@ -92,13 +80,9 @@ jobs:
       DISABLE_SQLITE_AUTO_BACKUP: "true"
       REQUIRE_AXE: "1"
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: "24"
-          cache: npm
       - run: npm ci
       - name: Cache Playwright browsers
         uses: actions/cache@v6.1.0

.github/workflows/nightly-schemathesis.yml

@@ -10,14 +10,12 @@ permissions:
 jobs:
   schemathesis:
     name: Schemathesis — OpenAPI contract fuzz (advisory)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     timeout-minutes: 30
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with: { node-version: "24", cache: npm }
       - run: npm ci
       - name: Build CLI bundle
         env:

.github/workflows/npm-publish.yml

@@ -62,7 +62,7 @@ jobs:
     # mid-"Creating an optimized production build" while v3.8.48 had still fit in 16min.
     # This job never runs on `pull_request`, so the fork-safety clause is always true here;
     # it is kept verbatim so the expression stays greppable against ci.yml.
-    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository)) && fromJSON('["self-hosted","omni-release"]') || 'ubuntu-latest' }}
+    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository)) && fromJSON('["self-hosted","omni-release"]') || 'ubuntu-26.04' }}
     permissions:
       actions: read # find + download the CI run's next-build artifact for this SHA
       contents: write # gh release upload (attach SBOM to the GitHub Release)
@@ -70,7 +70,7 @@ jobs:
       packages: write # publish to npm.pkg.github.com
     steps:
       - name: Checkout
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           # Need full tag history to compare against highest semver when
@@ -78,7 +78,7 @@ jobs:
           fetch-depth: 0
 
       - name: Setup Node.js
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
         with:
           node-version: ${{ env.NPM_PUBLISH_NODE_VERSION }}
           registry-url: https://registry.npmjs.org
@@ -339,20 +339,20 @@ jobs:
           echo "✅ Action finished for GitHub Packages"
 
   publish-opencode-plugin:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     permissions:
       contents: read
       id-token: write # npm provenance
     steps:
       - name: Checkout
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           fetch-depth: 0
           # Full history needed for auto-bump: git diff against previous release tag
 
       - name: Setup Node.js
-        uses: actions/setup-node@v7
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
         with:
           node-version: ${{ env.NPM_PUBLISH_NODE_VERSION }}
           registry-url: https://registry.npmjs.org

.github/workflows/opencode-plugin-ci.yml

@@ -26,16 +26,16 @@ defaults:
 jobs:
   test:
     name: Test (Node ${{ matrix.node }})
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     strategy:
       fail-fast: false
       matrix:
         node: ["22", "24"]
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
         with:
           node-version: ${{ matrix.node }}
           cache: npm
@@ -46,13 +46,13 @@ jobs:
 
   build:
     name: Build
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     needs: test
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
         with:
           node-version: "22"
           cache: npm

.github/workflows/opencode-provider-ci.yml

@@ -26,16 +26,16 @@ defaults:
 jobs:
   test:
     name: Test (Node ${{ matrix.node }})
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     strategy:
       fail-fast: false
       matrix:
         node: ["20", "22", "24"]
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
         with:
           node-version: ${{ matrix.node }}
           cache: npm
@@ -45,13 +45,13 @@ jobs:
 
   build:
     name: Build
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     needs: test
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
         with:
           node-version: "20"
           cache: npm

.github/workflows/quality.yml

@@ -25,20 +25,17 @@ jobs:
   # path filters share existence reasons: code / docs / i18n / workflow.
   changes:
     name: Change Classification
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     outputs:
       code: ${{ steps.classify.outputs.code }}
       docs: ${{ steps.classify.outputs.docs }}
       i18n: ${{ steps.classify.outputs.i18n }}
       workflow: ${{ steps.classify.outputs.workflow }}
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
           fetch-depth: 0
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
       - id: classify
         env:
           EVENT_NAME: ${{ github.event_name }}
@@ -61,19 +58,16 @@ jobs:
     name: Build (advisory)
     needs: changes
     if: ${{ github.event_name != 'pull_request' || ((github.event.pull_request.draft == false || startsWith(github.head_ref, 'mergify/merge-queue/')) && needs.changes.outputs.code == 'true') }}
-    # Dynamic runner — same fork-safe rule as ci.yml / fast-gates.
+    # Fork-safe fallback uses Ubuntu 26.04's bundled Node 24 without setup-node.
-    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository)) && fromJSON('["self-hosted","omni-release"]') || 'ubuntu-latest' }}
+    runs-on: ${{ (vars.USE_VPS_RUNNER == 'true' && (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository)) && fromJSON('["self-hosted","omni-release"]') || 'ubuntu-26.04' }}
     # #7307: advisory for the first week of release-PR runs; remove
     # continue-on-error after the production-build signal is stable.
     continue-on-error: true
     steps:
-      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
+      - run: node -e 'if (process.versions.node.split(".")[0] !== process.env.CI_NODE_VERSION) throw new Error("Expected Node " + process.env.CI_NODE_VERSION)'
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - uses: ./.github/actions/npm-ci-retry
       - run: npm run check:node-runtime
       - run: npm run build
@@ -88,15 +82,11 @@ jobs:
     name: Docs Gates (fast-path)
     needs: changes
     if: ${{ github.event_name != 'pull_request' || ((github.event.pull_request.draft == false || startsWith(github.head_ref, 'mergify/merge-queue/')) && (needs.changes.outputs.docs == 'true' || needs.changes.outputs.code == 'true')) }}
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - run: npm ci
       # One walk of src/app/api for openapi-routes + docs-symbols (both still fail independently).
       - run: npm run check:api-docs-refs
@@ -111,7 +101,7 @@ jobs:
     # Dynamic runner (same rule as ci.yml): use the self-hosted VPS pool only when the
     # release captain has USE_VPS_RUNNER=true AND this is not a fork PR (own-origin
     # branches only — a fork PR must never execute on the LAN runner). Var unset/false
-    # or a fork PR falls back to ubuntu-latest, so this is inert until the flag flips.
+    # or a fork PR falls back to ubuntu-26.04, so this is inert until the flag flips.
     # PINNED to hosted (gap 19). This job carried the USE_VPS_RUNNER expression, and that
     # expression was DEAD CONFIGURATION: across 160 quality.yml runs the job never once landed on
     # a self-hosted runner — every non-skipped sample is `GitHub Actions NNNN`. The classifier is
@@ -125,7 +115,7 @@ jobs:
     #
     # With this pinned, USE_VPS_RUNNER governs ONLY build-like jobs — one variable, one coherent
     # purpose. That is what gap 19 asked for; a second variable turned out to be unnecessary.
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     # tsx gates (known-symbols, route-guard-membership) import modules that open
     # SQLite on load; provide DB env so a fresh CI DB initializes cleanly.
     env:
@@ -133,14 +123,10 @@ jobs:
       API_KEY_SECRET: ci-lint-api-key-secret-long
       DISABLE_SQLITE_AUTO_BACKUP: "true"
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           fetch-depth: 0
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - run: npm ci
       - name: Restore ESLint file cache
         uses: actions/cache@v6
@@ -179,6 +165,81 @@ jobs:
       # Complexity + cognitive-complexity: ONE ESLint walk (both baselines still
       # enforced separately by ruleId). Avoids two cold tree walks on fast-path.
       - run: npm run check:complexity-ratchets
+      # ── G0 (trilho .50): gates do trilho A que faltavam no trilho B ──────────────
+      # The god-file refactor happens in PRs→release/**; without these, the release
+      # rail never sees a new import cycle, dead code, duplication or a security
+      # regression until the release PR to main. Deliberately NOT brought here:
+      # bundle-size (self-skips without a build — this rail's build job is advisory
+      # and uploads nothing, so it would be dead configuration) and the coverage
+      # run (fast-unit already runs the full suite; the coverage ratchet stays on
+      # the main rail via --allow-missing in lint-guard).
+      - run: npm run check:cycles
+      - run: npm run check:lockfile
+      - name: Duplication ratchet
+        run: npm run check:duplication
+      - name: Dead-code ratchet (knip)
+        run: npm run check:dead-code
+      - name: Type coverage ratchet
+        run: npm run check:type-coverage
+      - name: Compression budget ratchet
+        run: npm run check:compression-budget
+      # Security scanners — same hardened install as ci.yml quality-extended
+      # (gh release download = authenticated, 5000 req/hr; curl to api.github.com
+      # is rate-limited to 60/hr and silently no-ops when throttled). The blocking
+      # gates below SKIP (exit 0) when their binary is absent — only a measured
+      # regression vs config/quality/quality-baseline.json blocks.
+      - name: Install security scanners (gitleaks/osv/actionlint/zizmor/oasdiff)
+        continue-on-error: true
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          set +e
+          mkdir -p "$HOME/.local/bin"
+          # Ratchets compare scanner COUNTS across runs. Pin every auditor: a rule-set
+          # update must be an explicit PR that re-measures/rebaselines, never a random
+          # red (or green) caused by whatever "latest" served that morning.
+          GITLEAKS_VERSION=v8.30.1
+          OSV_SCANNER_VERSION=v2.3.8
+          ACTIONLINT_VERSION=v1.7.12
+          ZIZMOR_VERSION=1.25.2
+          OASDIFF_VERSION=v1.19.1
+          # gitleaks — pinned linux x64 tarball via gh (authed), extract binary
+          rm -rf /tmp/gl && mkdir -p /tmp/gl
+          gh release download "$GITLEAKS_VERSION" --repo gitleaks/gitleaks --pattern '*linux_x64.tar.gz' --dir /tmp/gl
+          tar -xzf /tmp/gl/*linux_x64.tar.gz -C "$HOME/.local/bin" gitleaks
+          # osv-scanner — pinned linux amd64 bare binary via gh (authed)
+          rm -rf /tmp/osv && mkdir -p /tmp/osv
+          gh release download "$OSV_SCANNER_VERSION" --repo google/osv-scanner --pattern '*linux_amd64' --dir /tmp/osv
+          install -m 0755 /tmp/osv/*linux_amd64 "$HOME/.local/bin/osv-scanner"
+          # actionlint — official installer from a pinned release tag (never main)
+          bash <(curl -fsSL "https://raw.githubusercontent.com/rhysd/actionlint/${ACTIONLINT_VERSION}/scripts/download-actionlint.bash") "$ACTIONLINT_VERSION" "$HOME/.local/bin"
+          # zizmor — pinned PyPI package (same version as ci.yml quality-extended)
+          pipx install "zizmor==$ZIZMOR_VERSION" || pip install --user "zizmor==$ZIZMOR_VERSION"
+          # oasdiff — pinned linux amd64 tarball via gh (authed), extract binary
+          rm -rf /tmp/oasd && mkdir -p /tmp/oasd
+          gh release download "$OASDIFF_VERSION" --repo Tufin/oasdiff --pattern '*linux_amd64.tar.gz' --dir /tmp/oasd
+          tar -xzf /tmp/oasd/*linux_amd64.tar.gz -C "$HOME/.local/bin" oasdiff
+          # ALWAYS export the bin dir (even if any step above failed)
+          echo "$HOME/.local/bin" >> "$GITHUB_PATH"
+          "$HOME/.local/bin/gitleaks" version || true
+          "$HOME/.local/bin/actionlint" -version || true
+          "$HOME/.local/bin/osv-scanner" --version || true
+          "$HOME/.local/bin/oasdiff" --version || true
+          zizmor --version || true
+      - name: Secret scan (gitleaks, ratchet, blocking)
+        run: npm run check:secrets -- --ratchet
+      - name: Vulnerability ratchet (osv-scanner, ratchet, blocking)
+        run: npm run check:vuln-ratchet -- --ratchet
+      - name: Workflow lint (actionlint+zizmor, ratchet, blocking)
+        run: npm run check:workflows -- --ratchet
+      # BASE_REF is read by the script from the env (never interpolated into a
+      # shell body) — workflow-injection-safe. actions/checkout fetches remote
+      # refs, not a local branch named github.base_ref, so prefix origin/ or this
+      # gate self-skips every PR with reason=base-unresolved.
+      - name: OpenAPI breaking-change (oasdiff, ratchet, blocking)
+        env:
+          BASE_REF: ${{ github.base_ref && format('origin/{0}', github.base_ref) || '' }}
+        run: npm run check:openapi-breaking -- --ratchet
       - name: Typecheck (core)
         run: npm run typecheck:core
       # #7033: dashboard-scoped typecheck gate — src/app/(dashboard) TSX is not
@@ -204,7 +265,7 @@ jobs:
       # selector returns __RUN_ALL__ — full-suite authority is the parallel
       # `fast-unit` 4-shard job (test:unit:ci:shard; was 2-shard, #6781), NOT an
       # unsharded re-run here. Stacking unsharded test:unit:ci on top of fast-unit
-      # doubled wall time (~16 min extra on ubuntu-latest) without extra coverage.
+      # doubled wall time (~16 min extra on the hosted runner) without extra coverage.
       #
       # BLOCKING for the *impacted subset* (flipped 2026-06-17). Fail-safe full
       # coverage remains required via `Unit Tests fast-path` (fast-unit).
@@ -268,36 +329,15 @@ jobs:
           if-no-files-found: ignore
           retention-days: 30
 
-  fast-vitest:
+      # Share fast-gates' checkout + npm ci instead of spending ~80s preparing a
-    name: Vitest (fast-path)
+      # separate runner for a ~13s Vitest invocation. !cancelled() preserves the
-    needs: changes
+      # independent test signal when an earlier fast gate fails.
-    if: ${{ github.event_name != 'pull_request' || ((github.event.pull_request.draft == false || startsWith(github.head_ref, 'mergify/merge-queue/')) && needs.changes.outputs.code == 'true') }}
+      - name: Vitest
-    # Dynamic runner — see fast-gates (own-origin + flag; fork/unset → ubuntu-latest).
+        if: ${{ !cancelled() }}
-    # PINNED to hosted, deliberately not on the USE_VPS_RUNNER switch (gap 19). One variable
+        run: npm run test:vitest -- --reporter=default --reporter=junit --outputFile.junit=trunk-junit/vitest-fastpath.xml
-    # governed the build and the test jobs, which want OPPOSITE machines: the build needs the
+      # WS5.2/5.3: JUnit feeds Trunk Flaky Tests — the fast path runs on every PR.
-    # .113's RAM, the tests need the hosted runner's link. Measured on 2026-07-29 —
+      # Advisory upload, own-origin only.
-    # actions/setup-node took 20m06s on .113 with 4 concurrent runners versus 16s hosted (npm
+      - name: Upload Vitest results to Trunk (advisory)
-    # cache restore saturating the link), while the tests themselves tied, 2m54 vs 2m31. So
-    # self-hosted is strictly worse here and there is nothing to configure.
-    runs-on: ubuntu-latest
-    env:
-      JWT_SECRET: ci-lint-secret-with-sufficient-length-for-validation
-      API_KEY_SECRET: ci-lint-api-key-secret-long
-      DISABLE_SQLITE_AUTO_BACKUP: "true"
-    steps:
-      - uses: actions/checkout@v7
-        with:
-          persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
-      - run: npm ci
-      # WS5.2/5.3: JUnit feeds Trunk Flaky Tests — the fast-path runs on EVERY PR,
-      # which is where flaky-detection volume actually comes from (ci.yml's heavy
-      # jobs only run on the release PR). Advisory upload, own-origin only.
-      - run: npm run test:vitest -- --reporter=default --reporter=junit --outputFile.junit=trunk-junit/vitest-fastpath.xml
-      - name: Upload test results to Trunk (advisory)
         if: ${{ always() && (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository) }}
         continue-on-error: true
         uses: trunk-io/analytics-uploader@385f1ccdf345b4532dc4b6c665dd432b702b8e28 # v2.1.2
@@ -310,9 +350,9 @@ jobs:
     name: Unit Tests fast-path (${{ matrix.shard }}/4)
     needs: changes
     if: ${{ github.event_name != 'pull_request' || ((github.event.pull_request.draft == false || startsWith(github.head_ref, 'mergify/merge-queue/')) && needs.changes.outputs.code == 'true') }}
-    # Dynamic runner — see fast-gates (own-origin + flag; fork/unset → ubuntu-latest).
+    # Dynamic runner — see fast-gates (own-origin + flag; fork/unset → ubuntu-26.04).
     # This is the heaviest fast-path job; 4-way sharding (was 2, #6781) halves the
-    # critical path again (~8.5min → ~4.5min on ubuntu-latest; ~2min on the 8-slot
+    # critical path again (~8.5min → ~4.5min hosted; ~2min on the 8-slot
     # runner box). Node's native --test-shard=N/total takes any denominator — only
     # this matrix and the TEST_SHARD env below encode the shard count.
     # PINNED to hosted, deliberately not on the USE_VPS_RUNNER switch (gap 19). One variable
@@ -321,7 +361,7 @@ jobs:
     # actions/setup-node took 20m06s on .113 with 4 concurrent runners versus 16s hosted (npm
     # cache restore saturating the link), while the tests themselves tied, 2m54 vs 2m31. So
     # self-hosted is strictly worse here and there is nothing to configure.
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     strategy:
       fail-fast: false
       matrix:
@@ -331,13 +371,9 @@ jobs:
       API_KEY_SECRET: ci-lint-api-key-secret-long
       DISABLE_SQLITE_AUTO_BACKUP: "true"
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - run: npm ci
       # QW-d: fonte única — o mesmo npm script do CI pesado/local. Fecha dois drifts do
       # comando inline antigo: os dirs `memory` e `usage` estavam FORA do glob (testes
@@ -362,16 +398,18 @@ jobs:
     name: No new ESLint warnings
     needs: changes
     if: ${{ github.event_name != 'pull_request' || ((github.event.pull_request.draft == false || startsWith(github.head_ref, 'mergify/merge-queue/')) && needs.changes.outputs.code == 'true') }}
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     continue-on-error: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.repo.fork == true }}
+    # G0 (trilho .50): security-events:read lets the CodeQL ratchet below read open
+    # code-scanning alerts via `gh api .../code-scanning/alerts` (same as ci.yml's
+    # quality-gate job). contents: read keeps checkout working.
+    permissions:
+      contents: read
+      security-events: read
     steps:
-      - uses: actions/checkout@v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
-      - uses: actions/setup-node@v7
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - run: npm ci
       - name: Restore ESLint file cache
         uses: actions/cache@v6
@@ -385,6 +423,29 @@ jobs:
       - name: ESLint (baseline congelado — warning novo = vermelho)
         # lint:json writes the report; --max-warnings 0 keeps no-new-warnings policy.
         run: npm run lint:json -- --max-warnings 0
+      # ── G0 (trilho .50): motor de ratchet também no trilho B ─────────────────────
+      # This job just wrote .artifacts/eslint-results.json — collect-metrics prefers
+      # that file, so the ratchet engine lands here at ZERO extra ESLint cost (one
+      # inventory, two consumers; same reason ci.yml chains lint → quality-gate).
+      # The coverage-report artifact does not exist on this rail, so both ratchet
+      # invocations run --allow-missing: coverage.* metrics skip gracefully while
+      # the deterministic ones (eslint / openapi-coverage / i18n-ui) stay BLOCKING.
+      # Coverage authority remains on the main rail (ci.yml test-coverage → quality-gate).
+      - run: npm run quality:collect
+      - name: Ratchet check (blocking)
+        run: node scripts/quality/check-quality-ratchet.mjs --allow-missing --summary .artifacts/quality-ratchet.md
+      - name: Require-tighten (blocking)
+        run: node scripts/quality/check-quality-ratchet.mjs --allow-missing --require-tighten
+      # CodeQL alerts ratchet — same semantics as ci.yml quality-gate: exits 1 ONLY
+      # on a real regression (open alerts > baseline in quality-baseline.json);
+      # a measurement failure (gh/auth/api) self-skips with exit 0.
+      - name: CodeQL alerts ratchet (blocking)
+        run: npm run check:codeql-ratchet
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+      - name: Append ratchet summary
+        if: always()
+        run: cat .artifacts/quality-ratchet.md >> "$GITHUB_STEP_SUMMARY" || true
 
   # Merge-integrity: pega no PR os dois vazamentos crônicos de merge que hoje só
   # explodem na release-PR. (1) CHANGELOG-eat — o auto-resolve do merge come
@@ -401,21 +462,17 @@ jobs:
     name: Merge integrity (changelog + generated skills)
     # Always on non-draft PRs — CHANGELOG/skills can break on docs-only merges too.
     if: ${{ github.event_name != 'pull_request' || (github.event.pull_request.draft == false || startsWith(github.head_ref, 'mergify/merge-queue/')) }}
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     continue-on-error: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.repo.fork == true }}
     env:
       JWT_SECRET: ci-lint-secret-with-sufficient-length-for-validation
       API_KEY_SECRET: ci-lint-api-key-secret-long
       DISABLE_SQLITE_AUTO_BACKUP: "true"
     steps:
-      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           fetch-depth: 0
           persist-credentials: false
-      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v6
-        with:
-          node-version: ${{ env.CI_NODE_VERSION }}
-          cache: npm
       - run: npm ci
       - name: CHANGELOG integrity (nenhum bullet da base pode sumir no merge-result)
         run: npm run check:changelog-integrity

.github/workflows/scorecard.yml

@@ -11,7 +11,7 @@ permissions: read-all
 jobs:
   analysis:
     name: Scorecard analysis
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     permissions:
       # security-events: write removed — Scorecard findings are advisory and no longer
       # uploaded to the code-scanning Security tab (they are supply-chain/posture scores,
@@ -21,7 +21,7 @@ jobs:
       contents: read
     steps:
       - name: Checkout
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
 

.github/workflows/semgrep.yml

@@ -14,7 +14,7 @@ concurrency:
   cancel-in-progress: true
 jobs:
   semgrep:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     container:
       image: semgrep/semgrep
     steps:

.github/workflows/wiki-sync.yml

@@ -34,15 +34,10 @@ concurrency:
 jobs:
   sync-wiki:
     name: Sync wiki with docs
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-26.04
     steps:
       - name: Checkout repo
-        uses: actions/checkout@v7
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
-
-      - name: Setup Node
-        uses: actions/setup-node@v7
-        with:
-          node-version: "24"
 
       - name: Clone wiki
         env:

.gitignore

@@ -72,6 +72,7 @@ yarn-error.log*
 # env files (can opt-in for committing if needed)
 .env*
 !.env.example
+!.env.devin-bridge.example
 !.env.homolog.example
 # Provider API keys (never commit)
 *.api-key
@@ -171,7 +172,6 @@ config/quality/test-impact-map.json
 # GitNexus local index
 .gitnexus
 .worktrees
-bin/omniroute.mjs
 
 # Consistent with .dockerignore / .npmignore
 .omc/
@@ -201,12 +201,17 @@ scripts/i18n/_pending-keys.json
 .codegraph/
 
 # Fumadocs generated source
-.source/
+/.source/
+
+# Temporary local worktrees used to build unpublished npm tarballs
+/.deploy-build-*/
 
 # AI agent local settings and configs
 .agents/
 .antigravitycli/
 .claude/
+!tests/fixtures/devin-bridge/e2e-workspace/.claude/
+!tests/fixtures/devin-bridge/e2e-workspace/.claude/**
 
 # PR Reviews and local feedback files
 pr_reviews*.json
@@ -243,6 +248,8 @@ _artifacts/          # release-green artifacts
 
 # CI/local quality artifacts (eslint-results.json, quality-ratchet.md, etc.)
 .artifacts/
+# Isolated Devin bridge workspaces, evidence, and test databases
+.sandbox/
 
 # Homologation E2E suite (npm run homolog) — real-environment credentials + report output
 .env.homolog
@@ -250,3 +257,4 @@ tests/homolog/.auth/
 tests/homolog/ui/.auth/
 homolog-report/
 docker-compose.yml.bak
+.playwright-cli/

.source/dynamic.ts

@@ -1,8 +0,0 @@
-// @ts-nocheck
-import { dynamic } from 'fumadocs-mdx/runtime/dynamic';
-import * as Config from '../source.config';
-
-const create = await dynamic<typeof Config, import("fumadocs-mdx/runtime/types").InternalTypeConfig & {
-  DocData: {
-  }
-}>(Config, {"configPath":"source.config.ts","environment":"next","outDir":".source"}, {"doc":{"passthroughs":["extractedReferences"]}});

.source/source.config.mjs

@@ -1,22 +0,0 @@
-// source.config.ts
-import { defineDocs, defineConfig } from "fumadocs-mdx/config";
-var docs = defineDocs({
-  dir: "docs",
-  docs: {
-    files: [
-      "./architecture/**/*.md",
-      "./guides/**/*.md",
-      "./reference/**/*.md",
-      "./frameworks/**/*.md",
-      "./routing/**/*.md",
-      "./security/**/*.md",
-      "./compression/**/*.md",
-      "./ops/**/*.md"
-    ]
-  }
-});
-var source_config_default = defineConfig();
-export {
-  source_config_default as default,
-  docs
-};

@omniroute/opencode-plugin/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "@omniroute/opencode-plugin",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@omniroute/opencode-plugin",
-      "version": "0.2.0",
+      "version": "0.2.1",
       "license": "MIT",
       "dependencies": {
         "zod": "^4.4.3"

@omniroute/openhands-plugin/.gitignore

@@ -0,0 +1,5 @@
+node_modules
+dist
+*.log
+.DS_Store
+.env

@omniroute/openhands-plugin/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 OmniRoute contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

@omniroute/openhands-plugin/README.md

@@ -0,0 +1,118 @@
+# @omniroute/openhands-plugin
+
+OpenHands integration for the **OmniRoute AI Gateway**. Generates the OpenHands
+environment and Docker config that wires an OpenHands agent-server to a running
+OmniRoute instance — with the integration gotchas already handled.
+
+## Why
+
+Running OpenHands against OmniRoute directly hits several wall:
+
+1. **Model name mismatch** — OpenHands sends `model: "deepseek-chat"`, OmniRoute
+   uses provider-prefixed IDs (`ds/deepseek-v4-flash`) or combos.
+2. **Python 3.13 sandbox** — `socket.socketpair()` fails under Docker's default
+   seccomp profile; the agent-server needs `privileged: true`.
+3. **Host reachability** — the sandbox can't resolve `localhost` to the OmniRoute
+   host; needs `host.docker.internal:host-gateway`.
+4. **Lost state** — conversations die with the container unless
+   `OH_PERSISTENCE_DIR` is a host volume.
+5. **CORS** — the dashboard origin can't reach agent-server unless
+   `PERMITTED_CORS_ORIGINS` allows it.
+
+This plugin encodes all of that into one command.
+
+## Install
+
+```bash
+npm install -g @omniroute/openhands-plugin
+# or: npx @omniroute/openhands-plugin ...
+```
+
+## Quick start
+
+Generate the OpenHands `.env`:
+
+```bash
+omniroute-openhands env \
+  --api-key sk-... \
+  --model deepseek-chat \
+  --url http://192.168.3.106:20128
+```
+
+Generate a `docker-compose.yml` service:
+
+```bash
+omniroute-openhands compose \
+  --api-key sk-... \
+  --model glm-5.2 \
+  --persistence-dir /Users/me/.openhands-state \
+  --cors-origins http://100.73.44.17:3000
+```
+
+Or a plain `docker run`:
+
+```bash
+omniroute-openhands docker-run \
+  --api-key sk-... \
+  --model vivanta-core \
+  --persistence-dir /Users/me/.openhands-state
+```
+
+## Commands
+
+| Command | Description |
+|---------|-------------|
+| `env` | Print OpenHands `.env` contents |
+| `compose` | Print a Docker Compose service block |
+| `docker-run` | Print a full `docker run` command |
+| `models` | Print the default OpenHands → OmniRoute model map |
+
+### Common options
+
+| Flag | Description | Default |
+|------|-------------|---------|
+| `--api-key` | OmniRoute API key (`sk-...`) | — |
+| `--model` | OpenHands model name or OmniRoute combo | — |
+| `--url` | OmniRoute base URL | `http://localhost:20128` |
+| `--persistence-dir` | Host dir for conversation state | `.openhands-state` |
+| `--cors-origins` | Comma-separated allowed origins | `localhost:3000,3001` |
+| `--sandbox-image` | OpenHands sandbox base image | — |
+
+## Model mapping
+
+OpenHands-friendly names are mapped to OmniRoute IDs/combo names:
+
+| OpenHands sends | OmniRoute resolves to |
+|-----------------|----------------------|
+| `deepseek-chat` | `ds/deepseek-v4-flash` |
+| `deepseek-reasoner` | `ds/deepseek-v4-pro` |
+| `glm-5.2` | `nvidia/z-ai/glm-5.2` |
+| `gpt-4o` | `openai/gpt-4o` |
+| `claude-sonnet-4.5` | `anthropic/claude-sonnet-4.5` |
+| ... | ... |
+
+Or just pass an OmniRoute combo name (e.g. `--model vivanta-core`) — the Model
+Alias Resolver and combo router accept it directly.
+
+## Library usage
+
+```ts
+import {
+  buildOpenHandsEnv,
+  serializeOpenHandsEnv,
+  buildOpenHandsCompose,
+  resolveOpenHandsModel,
+} from "@omniroute/openhands-plugin";
+
+const env = buildOpenHandsEnv({
+  apiKey: "sk-...",
+  model: resolveOpenHandsModel("deepseek-chat"),
+  omnirouteUrl: "http://localhost:20128",
+  persistenceDir: "/Users/me/.openhands-state",
+});
+console.log(serializeOpenHandsEnv(env));
+```
+
+## License
+
+MIT — same as OmniRoute.

@omniroute/openhands-plugin/package.json

@@ -0,0 +1,79 @@
+{
+  "name": "@omniroute/openhands-plugin",
+  "version": "0.1.0",
+  "description": "OpenHands integration for the OmniRoute AI Gateway. Generates OpenHands env + Docker Compose config (model mapping, sandbox, CORS, persistence) so OpenHands agents talk to OmniRoute out of the box.",
+  "type": "module",
+  "main": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "bin": {
+    "omniroute-openhands": "./dist/cli.js"
+  },
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "import": "./dist/index.js"
+    },
+    "./env": {
+      "types": "./dist/env.d.ts",
+      "import": "./dist/env.js"
+    },
+    "./docker": {
+      "types": "./dist/docker.d.ts",
+      "import": "./dist/docker.js"
+    },
+    "./model-map": {
+      "types": "./dist/model-map.d.ts",
+      "import": "./dist/model-map.js"
+    }
+  },
+  "files": [
+    "dist",
+    "README.md",
+    "LICENSE"
+  ],
+  "scripts": {
+    "build": "tsup",
+    "clean": "rm -rf dist",
+    "test": "node --import tsx/esm --test tests/env.test.ts tests/model-map.test.ts tests/docker.test.ts",
+    "prepublishOnly": "npm run clean && npm run build && npm test"
+  },
+  "keywords": [
+    "omniroute",
+    "openhands",
+    "open-hands",
+    "openhands-plugin",
+    "openai-compatible",
+    "docker",
+    "agent"
+  ],
+  "author": "OmniRoute contributors",
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/diegosouzapw/OmniRoute.git",
+    "directory": "@omniroute/openhands-plugin"
+  },
+  "bugs": {
+    "url": "https://github.com/diegosouzapw/OmniRoute/issues"
+  },
+  "homepage": "https://github.com/diegosouzapw/OmniRoute/tree/main/%40omniroute/openhands-plugin#readme",
+  "engines": {
+    "node": ">=18.0.0"
+  },
+  "peerDependencies": {
+    "@omniroute/open-sse": "*"
+  },
+  "peerDependenciesMeta": {
+    "@omniroute/open-sse": {
+      "optional": true
+    }
+  },
+  "publishConfig": {
+    "access": "public"
+  },
+  "devDependencies": {
+    "@types/node": "^22.19.19",
+    "tsup": "^8.5.1",
+    "tsx": "^4.22.3"
+  }
+}

@omniroute/openhands-plugin/src/cli.ts

@@ -0,0 +1,104 @@
+#!/usr/bin/env node
+/**
+ * @omniroute/openhands-plugin CLI — generate OpenHands .env / Docker config
+ * for a running OmniRoute instance.
+ *
+ * Usage:
+ *   omniroute-openhands env   --api-key sk-... --model deepseek-chat [--url http://localhost:20128]
+ *   omniroute-openhands compose --api-key sk-... --model deepseek-chat [--persistence-dir /path]
+ *   omniroute-openhands docker-run --api-key sk-... --model deepseek-chat
+ *   omniroute-openhands models  (print the default model map)
+ */
+import { buildOpenHandsEnv, serializeOpenHandsEnv } from "./env.ts";
+import { buildOpenHandsCompose, buildOpenHandsDockerRun } from "./docker.ts";
+import { DEFAULT_OPENHANDS_MODEL_MAP } from "./model-map.ts";
+
+function parseArgs(argv: string[]): Record<string, string> {
+  const out: Record<string, string> = {};
+  for (let i = 0; i < argv.length; i++) {
+    const arg = argv[i];
+    if (!arg.startsWith("--")) continue;
+    const key = arg.slice(2);
+    const next = argv[i + 1];
+    if (next !== undefined && !next.startsWith("--")) {
+      out[key] = next;
+      i++;
+    } else {
+      out[key] = "true";
+    }
+  }
+  return out;
+}
+
+function requireArgs(args: Record<string, string>, names: string[]): void {
+  for (const name of names) {
+    if (!args[name]) {
+      console.error(`Missing required --${name}`);
+      process.exit(2);
+    }
+  }
+}
+
+const [cmd, ...rest] = process.argv.slice(2);
+const args = parseArgs(rest);
+
+switch (cmd) {
+  case "env": {
+    requireArgs(args, ["api-key", "model"]);
+    const env = buildOpenHandsEnv({
+      apiKey: args["api-key"],
+      model: args.model,
+      omnirouteUrl: args.url,
+      persistenceDir: args["persistence-dir"],
+      corsOrigins: args["cors-origins"]?.split(","),
+    });
+    process.stdout.write(serializeOpenHandsEnv(env));
+    break;
+  }
+  case "compose": {
+    requireArgs(args, ["api-key", "model"]);
+    process.stdout.write(
+      buildOpenHandsCompose({
+        apiKey: args["api-key"],
+        model: args.model,
+        omnirouteUrl: args.url,
+        persistenceDir: args["persistence-dir"] ?? ".openhands-state",
+        corsOrigins: args["cors-origins"]?.split(","),
+        sandboxBaseImage: args["sandbox-image"],
+      })
+    );
+    break;
+  }
+  case "docker-run": {
+    requireArgs(args, ["api-key", "model"]);
+    process.stdout.write(
+      buildOpenHandsDockerRun({
+        apiKey: args["api-key"],
+        model: args.model,
+        omnirouteUrl: args.url,
+        persistenceDir: args["persistence-dir"] ?? ".openhands-state",
+        corsOrigins: args["cors-origins"]?.split(","),
+        sandboxBaseImage: args["sandbox-image"],
+      })
+    );
+    break;
+  }
+  case "models": {
+    for (const [name, target] of Object.entries(DEFAULT_OPENHANDS_MODEL_MAP)) {
+      process.stdout.write(`${name}\t->\t${target}\n`);
+    }
+    break;
+  }
+  default:
+    console.error(
+      "Usage: omniroute-openhands <env|compose|docker-run|models> [options]\n" +
+        "Options:\n" +
+        "  --api-key <sk-...>         OmniRoute API key (required for env/compose/docker-run)\n" +
+        "  --model <name>             OpenHands model name or OmniRoute combo\n" +
+        "  --url <base>               OmniRoute URL (default http://localhost:20128)\n" +
+        "  --persistence-dir <path>   Host dir for conversation state\n" +
+        "  --cors-origins <a,b,...>   Allowed CORS origins\n" +
+        "  --sandbox-image <image>    OpenHands sandbox base image"
+    );
+    process.exit(1);
+}

@omniroute/openhands-plugin/src/docker.ts

@@ -0,0 +1,92 @@
+/**
+ * OpenHands agent-server Docker Compose generator for OmniRoute.
+ *
+ * Bakes in the integration fixes that were needed to run OpenHands against
+ * OmniRoute reliably:
+ *   - `privileged: true`       — Python 3.13 socket.socketpair() needs it under
+ *                                Docker's default seccomp profile
+ *   - `extra_hosts`            — host.docker.internal → host-gateway so the
+ *                                sandbox can reach OmniRoute on the host
+ *   - host volume for OH_PERSISTENCE_DIR so conversations survive `docker rm`
+ *   - PERMITTED_CORS_ORIGINS   — allow the dashboard origin to hit agent-server
+ */
+
+export interface OpenHandsDockerOptions {
+  /** Agent-server image (default: the official OpenHands runtime image). */
+  image?: string;
+  /** Container name (default: openhands-agent). */
+  containerName?: string;
+  /** Model name to pass via LLM_MODEL. */
+  model: string;
+  /** OmniRoute API key. */
+  apiKey: string;
+  /** OmniRoute base URL reachable from the sandbox (default http://localhost:20128). */
+  omnirouteUrl?: string;
+  /** Host directory for OH_PERSISTENCE_DIR (must match env.ts persistenceDir). */
+  persistenceDir: string;
+  /** CORS origins to permit. */
+  corsOrigins?: string[];
+  /** Sandbox base image (defaults to OpenHands default). */
+  sandboxBaseImage?: string;
+  /** Set true to use host networking instead of extra_hosts. */
+  hostNetwork?: boolean;
+}
+
+export function buildOpenHandsCompose(opts: OpenHandsDockerOptions): string {
+  const image = opts.image ?? "docker.all-hands.dev/all-hands-ai/openhands:latest";
+  const containerName = opts.containerName ?? "openhands-agent";
+  const omnirouteHost = (opts.omnirouteUrl ?? "http://localhost:20128").replace(/\/+$/, "");
+  const cors =
+    opts.corsOrigins && opts.corsOrigins.length > 0
+      ? opts.corsOrigins
+      : ["http://localhost:3000", "http://localhost:3001"];
+
+  const lines: string[] = [];
+  lines.push(`services:`);
+  lines.push(`  openhands:`);
+  lines.push(`    image: ${image}`);
+  lines.push(`    container_name: ${containerName}`);
+  lines.push(`    privileged: true`);
+  lines.push(`    environment:`);
+  lines.push(`      LLM_MODEL: "${opts.model}"`);
+  lines.push(`      LLM_BASE_URL: "${omnirouteHost}/v1"`);
+  lines.push(`      LLM_API_KEY: "${opts.apiKey}"`);
+  lines.push(`      OH_PERSISTENCE_DIR: "/opt/.openhands-state"`);
+  lines.push(`      PERMITTED_CORS_ORIGINS: "${cors.join(",")}"`);
+  if (opts.sandboxBaseImage) {
+    lines.push(`      SANDBOX_BASE_IMAGE: "${opts.sandboxBaseImage}"`);
+  }
+  lines.push(`    volumes:`);
+  lines.push(`      - ${opts.persistenceDir}:/opt/.openhands-state`);
+  lines.push(`    extra_hosts:`);
+  lines.push(`      - "host.docker.internal:host-gateway"`);
+
+  return lines.join("\n") + "\n";
+}
+
+/**
+ * docker run equivalent of {@link buildOpenHandsCompose} — returns the full
+ * `docker run` command line.
+ */
+export function buildOpenHandsDockerRun(opts: OpenHandsDockerOptions): string {
+  const image = opts.image ?? "docker.all-hands.dev/all-hands-ai/openhands:latest";
+  const omnirouteHost = (opts.omnirouteUrl ?? "http://localhost:20128").replace(/\/+$/, "");
+  const cors =
+    opts.corsOrigins && opts.corsOrigins.length > 0
+      ? opts.corsOrigins
+      : ["http://localhost:3000", "http://localhost:3001"];
+
+  const parts = [
+    "docker run",
+    "--privileged",
+    "--add-host host.docker.internal:host-gateway",
+    `-e LLM_MODEL="${opts.model}"`,
+    `-e LLM_BASE_URL="${omnirouteHost}/v1"`,
+    `-e LLM_API_KEY="${opts.apiKey}"`,
+    `-e OH_PERSISTENCE_DIR=/opt/.openhands-state`,
+    `-e PERMITTED_CORS_ORIGINS="${cors.join(",")}"`,
+    `-v "${opts.persistenceDir}:/opt/.openhands-state"`,
+    image,
+  ];
+  return parts.join(" ") + "\n";
+}

@omniroute/openhands-plugin/src/env.ts

@@ -0,0 +1,63 @@
+/**
+ * OpenHands `.env` generator for the OmniRoute AI Gateway.
+ *
+ * Produces the OpenHands environment that points an OpenHands agent-server at
+ * a running OmniRoute instance and fixes the integration gotchas found in the
+ * field:
+ *   - LLM_MODEL   — OpenHands-friendly model name → OmniRoute model/combo
+ *   - LLM_BASE_URL — OmniRoute OpenAI-compatible endpoint
+ *   - LLM_API_KEY  — OmniRoute key (sk-...)
+ *   - OH_PERSISTENCE_DIR — host-mounted SQLite/conversation persistence
+ *   - PERMITTED_CORS_ORIGINS — allow the dashboard origin to reach agent-server
+ */
+
+export interface OpenHandsEnvOptions {
+  /** OmniRoute base URL as seen from the agent-server (default localhost:20128). */
+  omnirouteUrl?: string;
+  /** OmniRoute API key (sk-...). */
+  apiKey: string;
+  /** OpenHands model name (e.g. "deepseek-chat") or OmniRoute combo/model. */
+  model: string;
+  /** Host directory for OH_PERSISTENCE_DIR (default: current dir + .openhands-state). */
+  persistenceDir?: string;
+  /** CORS origins that must reach the agent-server (default dashboard origin + localhost). */
+  corsOrigins?: string[];
+  /** Optional OpenHands sandbox base image. */
+  sandboxBaseImage?: string;
+}
+
+export function buildOpenHandsEnv(opts: OpenHandsEnvOptions): Record<string, string> {
+  const omnirouteHost = (opts.omnirouteUrl ?? "http://localhost:20128").replace(/\/+$/, "");
+  const persistence = opts.persistenceDir ?? `${process.cwd()}/.openhands-state`;
+  const cors =
+    opts.corsOrigins && opts.corsOrigins.length > 0
+      ? opts.corsOrigins
+      : ["http://localhost:3000", "http://localhost:3001"];
+
+  const env: Record<string, string> = {
+    LLM_MODEL: opts.model,
+    LLM_BASE_URL: `${omnirouteHost}/v1`,
+    LLM_API_KEY: opts.apiKey,
+    OH_PERSISTENCE_DIR: persistence,
+    PERMITTED_CORS_ORIGINS: cors.join(","),
+  };
+
+  if (opts.sandboxBaseImage) {
+    env.SANDBOX_BASE_IMAGE = opts.sandboxBaseImage;
+  }
+
+  return env;
+}
+
+/**
+ * Serialize the env record to `.env` file content (KEY=VALUE lines).
+ * Values are not quoted unless they contain whitespace or `#`.
+ */
+export function serializeOpenHandsEnv(env: Record<string, string>): string {
+  const lines: string[] = [];
+  for (const [key, value] of Object.entries(env)) {
+    const needsQuotes = /[\s#]/.test(value);
+    lines.push(needsQuotes ? `${key}="${value}"` : `${key}=${value}`);
+  }
+  return lines.join("\n") + "\n";
+}

@omniroute/openhands-plugin/src/index.ts

@@ -0,0 +1,18 @@
+/**
+ * @omniroute/openhands-plugin — OpenHands integration for the OmniRoute AI Gateway.
+ *
+ * Generates the OpenHands environment and Docker Compose / docker run config
+ * that wires an OpenHands agent-server to a running OmniRoute instance:
+ * model mapping, sandbox privileges, host-gateway networking, persistent
+ * conversation state and CORS.
+ */
+export { buildOpenHandsEnv, serializeOpenHandsEnv } from "./env.ts";
+export type { OpenHandsEnvOptions } from "./env.ts";
+export { buildOpenHandsCompose, buildOpenHandsDockerRun } from "./docker.ts";
+export type { OpenHandsDockerOptions } from "./docker.ts";
+export {
+  DEFAULT_OPENHANDS_MODEL_MAP,
+  resolveOpenHandsModel,
+  buildOpenHandsModel,
+} from "./model-map.ts";
+export type { OpenHandsModelMap } from "./model-map.ts";

@omniroute/openhands-plugin/src/model-map.ts

@@ -0,0 +1,64 @@
+/**
+ * OpenHands → OmniRoute model mapping.
+ *
+ * OpenHands sends `model: "<LLM_MODEL>"` and expects the OpenAI-compatible
+ * endpoint to accept that exact string. OmniRoute uses provider-prefixed
+ * model IDs (`ds/deepseek-v4-flash`) and combo names. This module maps
+ * common OpenHands-friendly names to the OmniRoute model/combo they should
+ * resolve to, and back-fills the `LLM_MODEL` value for OpenHands.
+ */
+
+export interface OpenHandsModelMap {
+  /** OpenHands-friendly model name (e.g. "deepseek-chat") */
+  [openHandsName: string]: string;
+}
+
+/**
+ * Default mapping for the model names OpenHands and the broader ecosystem
+ * commonly send. Values are OmniRoute model IDs or combo names. Extend or
+ * override via {@link resolveOpenHandsModel}.
+ */
+export const DEFAULT_OPENHANDS_MODEL_MAP: OpenHandsModelMap = Object.freeze({
+  // DeepSeek
+  "deepseek-chat": "ds/deepseek-v4-flash",
+  "deepseek-reasoner": "ds/deepseek-v4-pro",
+  // Claude / Anthropic
+  "claude-sonnet-4.5": "anthropic/claude-sonnet-4.5",
+  "claude-opus-4.1": "anthropic/claude-opus-4.1",
+  "claude-haiku-4.5": "anthropic/claude-haiku-4.5",
+  // GPT / OpenAI
+  "gpt-4o": "openai/gpt-4o",
+  "gpt-4o-mini": "openai/gpt-4o-mini",
+  "gpt-5": "openai/gpt-5",
+  // Gemini
+  "gemini-2.5-flash": "gemini/gemini-2.5-flash",
+  "gemini-2.5-pro": "gemini/gemini-2.5-pro",
+  // GLM / Z.AI (NVIDIA NIM free endpoint)
+  "glm-5.2": "nvidia/z-ai/glm-5.2",
+});
+
+/**
+ * Resolve the OmniRoute model ID for an OpenHands-friendly model name.
+ * Returns the input unchanged when no mapping exists (OmniRoute will try to
+ * resolve it as a literal model/combo).
+ */
+export function resolveOpenHandsModel(
+  openHandsModel: string,
+  map: OpenHandsModelMap = DEFAULT_OPENHANDS_MODEL_MAP
+): string {
+  if (!openHandsModel) return openHandsModel;
+  const mapped = map[openHandsModel];
+  return mapped ?? openHandsModel;
+}
+
+/**
+ * Build the `LLM_MODEL` value for OpenHands from an OmniRoute model ID/combo.
+ *
+ * OpenHands only surfaces the literal `LLM_MODEL` string in its UI, so for
+ * OmniRoute combos (e.g. "vivanta-core") that's already the right value.
+ * For provider-prefixed IDs, we return them as-is — the OmniRoute Model
+ * Alias Resolver accepts both the raw ID and aliases on the `/v1` endpoint.
+ */
+export function buildOpenHandsModel(omnirouteModelOrCombo: string): string {
+  return omnirouteModelOrCombo;
+}

@omniroute/openhands-plugin/tests/env.test.ts

@@ -0,0 +1,76 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { buildOpenHandsEnv, serializeOpenHandsEnv } from "../src/env.ts";
+import { resolveOpenHandsModel, buildOpenHandsModel } from "../src/model-map.ts";
+import { buildOpenHandsCompose, buildOpenHandsDockerRun } from "../src/docker.ts";
+
+test("buildOpenHandsEnv produces LLM vars pointing at OmniRoute", () => {
+  const env = buildOpenHandsEnv({
+    apiKey: "sk-test-123",
+    model: "deepseek-chat",
+    omnirouteUrl: "http://192.168.3.106:20128",
+    persistenceDir: "/opt/state",
+    corsOrigins: ["http://100.73.44.17:3000"],
+  });
+  assert.equal(env.LLM_MODEL, "deepseek-chat");
+  assert.equal(env.LLM_BASE_URL, "http://192.168.3.106:20128/v1");
+  assert.equal(env.LLM_API_KEY, "sk-test-123");
+  assert.equal(env.OH_PERSISTENCE_DIR, "/opt/state");
+  assert.equal(env.PERMITTED_CORS_ORIGINS, "http://100.73.44.17:3000");
+});
+
+test("serializeOpenHandsEnv quotes values with whitespace/#", () => {
+  const out = serializeOpenHandsEnv({ LLM_MODEL: "deepseek-chat", LLM_BASE_URL: "http://localhost:20128/v1" });
+  const lines = out.trim().split("\n");
+  assert.ok(lines.some((l) => l.startsWith("LLM_MODEL=deepseek-chat")));
+  assert.ok(lines.some((l) => l.startsWith("LLM_BASE_URL=http://localhost:20128/v1")));
+});
+
+test("resolveOpenHandsModel maps known names to OmniRoute IDs", () => {
+  assert.equal(resolveOpenHandsModel("deepseek-chat"), "ds/deepseek-v4-flash");
+  assert.equal(resolveOpenHandsModel("glm-5.2"), "nvidia/z-ai/glm-5.2");
+  assert.equal(resolveOpenHandsModel("gpt-4o"), "openai/gpt-4o");
+});
+
+test("resolveOpenHandsModel passes unknown names through unchanged", () => {
+  assert.equal(resolveOpenHandsModel("vivanta-core"), "vivanta-core");
+  assert.equal(resolveOpenHandsModel(""), "");
+});
+
+test("resolveOpenHandsModel accepts custom map overrides", () => {
+  const custom = { "my-alias": "nvidia/z-ai/glm-5.2" };
+  assert.equal(resolveOpenHandsModel("my-alias", custom), "nvidia/z-ai/glm-5.2");
+  assert.equal(resolveOpenHandsModel("deepseek-chat", custom), "deepseek-chat");
+});
+
+test("buildOpenHandsModel passes combo names through", () => {
+  assert.equal(buildOpenHandsModel("vivanta-core"), "vivanta-core");
+  assert.equal(buildOpenHandsModel("ds/deepseek-v4-flash"), "ds/deepseek-v4-flash");
+});
+
+test("buildOpenHandsCompose includes privileged, extra_hosts, volume, CORS", () => {
+  const compose = buildOpenHandsCompose({
+    apiKey: "sk-x",
+    model: "deepseek-chat",
+    persistenceDir: "/Users/me/.openhands-state",
+    corsOrigins: ["http://localhost:3000"],
+  });
+  assert.ok(compose.includes("privileged: true"), "privileged present");
+  assert.ok(compose.includes("host.docker.internal:host-gateway"), "host-gateway present");
+  assert.ok(compose.includes("/Users/me/.openhands-state"), "persistence volume present");
+  assert.ok(compose.includes("LLM_BASE_URL: \"http://localhost:20128/v1\""), "base url present");
+  assert.ok(compose.includes("PERMITTED_CORS_ORIGINS: \"http://localhost:3000\""), "cors present");
+});
+
+test("buildOpenHandsDockerRun produces a runnable docker command", () => {
+  const run = buildOpenHandsDockerRun({
+    apiKey: "sk-x",
+    model: "glm-5.2",
+    persistenceDir: "/opt/state",
+  });
+  assert.ok(run.startsWith("docker run"));
+  assert.ok(run.includes("--privileged"));
+  assert.ok(run.includes("--add-host host.docker.internal:host-gateway"));
+  assert.ok(run.includes("LLM_MODEL=\"glm-5.2\""));
+  assert.ok(run.includes("LLM_BASE_URL=\"http://localhost:20128/v1\""));
+});

@omniroute/openhands-plugin/tsconfig.json

@@ -0,0 +1,22 @@
+{
+  "compilerOptions": {
+    "target": "ES2022",
+    "module": "ESNext",
+    "moduleResolution": "Bundler",
+    "lib": ["ES2022"],
+    "types": ["node"],
+    "ignoreDeprecations": "6.0",
+    "strict": true,
+    "esModuleInterop": true,
+    "skipLibCheck": true,
+    "allowImportingTsExtensions": true,
+    "declaration": true,
+    "isolatedModules": true,
+    "forceConsistentCasingInFileNames": true,
+    "noUncheckedIndexedAccess": false,
+    "outDir": "dist",
+    "rootDir": "src"
+  },
+  "include": ["src/**/*.ts"],
+  "exclude": ["dist", "node_modules", "tests"]
+}

@omniroute/openhands-plugin/tsup.config.ts

@@ -0,0 +1,15 @@
+import { defineConfig } from "tsup";
+
+export default defineConfig({
+  entry: ["src/index.ts", "src/cli.ts"],
+  format: ["esm"],
+  dts: true,
+  clean: true,
+  sourcemap: false,
+  splitting: false,
+  treeshake: false,
+  target: "node18",
+  outDir: "dist",
+  minify: false,
+  cjsInterop: false,
+});

AMIT

@@ -1 +0,0 @@
-

CLAUDE.md

@@ -45,7 +45,7 @@ For full test matrix, see `CONTRIBUTING.md` → "Running Tests". For deep archit
 | Translators   | `open-sse/translator/`  | Format conversion (OpenAI↔Claude↔Gemini)                                                                                                                |
 | Transformer   | `open-sse/transformer/` | Responses API ↔ Chat Completions                                                                                                                        |
 | Services      | `open-sse/services/`    | Combo routing, rate limits, caching, etc                                                                                                                |
-| Database      | `src/lib/db/`           | SQLite domain modules (95 files, 110 migrations)                                                                                                        |
+| Database      | `src/lib/db/`           | SQLite domain modules (130 migrations)                                                                                                                  |
 | Domain/Policy | `src/domain/`           | Policy engine, cost rules, fallback logic                                                                                                               |
 | MCP Server    | `open-sse/mcp-server/`  | 104 tools (42 base + memory/skill/agentSkill/pool/notion/obsidian/gamification/plugin modules), 3 transports (stdio / SSE / Streamable HTTP), 31 scopes |
 | A2A Server    | `src/lib/a2a/`          | JSON-RPC 2.0 agent protocol                                                                                                                             |
@@ -72,7 +72,7 @@ Client → /v1/chat/completions (Next.js route)
 
 API routes follow a consistent pattern: `Route → CORS preflight → Zod body validation → Optional auth (extractApiKey/isValidApiKey) → API key policy enforcement → Handler delegation (open-sse)`. No global Next.js middleware — interception is route-specific.
 
-**Combo routing** (`open-sse/services/combo.ts`): 18 strategies (priority, weighted, fill-first, round-robin, p2c, random, least-used, cost-optimized, reset-aware, reset-window, headroom, strict-random, auto, lkgp, context-optimized, context-relay, fusion, pipeline). Each target calls `handleSingleModel()` which wraps `handleChatCore()` with per-target error handling and circuit breaker checks. The `fusion` strategy is the exception: it fans out to a panel of models in parallel, then a judge model synthesizes one final answer (`open-sse/services/fusion.ts`). See `docs/routing/AUTO-COMBO.md` for the 12-factor Auto-Combo scoring + the full strategy table and `docs/architecture/RESILIENCE_GUIDE.md` for the 3 resilience layers.
+**Combo routing** (`open-sse/services/combo.ts`): 19 public strategies (priority, weighted, fill-first, round-robin, p2c, random, least-used, cost-optimized, reset-aware, reset-window, headroom, strict-random, auto, lkgp, context-optimized, cache-optimized, context-relay, fusion, pipeline). Each target calls `handleSingleModel()` which wraps `handleChatCore()` with per-target error handling and circuit breaker checks. The `fusion` strategy is the exception: it fans out to a panel of models in parallel, then a judge model synthesizes one final answer (`open-sse/services/fusion.ts`). See `docs/routing/AUTO-COMBO.md` for the 13-factor Auto-Combo scoring + the full strategy table and `docs/architecture/RESILIENCE_GUIDE.md` for the 3 resilience layers.
 
 ---
 
@@ -332,7 +332,7 @@ For any non-trivial change, read the matching deep-dive first:
 | Repo navigation                               | `docs/architecture/REPOSITORY_MAP.md`                   |
 | Architecture                                  | `docs/architecture/ARCHITECTURE.md`                     |
 | Engineering reference                         | `docs/architecture/CODEBASE_DOCUMENTATION.md`           |
-| Auto-Combo (12-factor scoring, 18 strategies) | `docs/routing/AUTO-COMBO.md`                            |
+| Auto-Combo (13-factor scoring, 19 strategies) | `docs/routing/AUTO-COMBO.md`                            |
 | Resilience (3 mechanisms)                     | `docs/architecture/RESILIENCE_GUIDE.md`                 |
 | Reasoning replay                              | `docs/routing/REASONING_REPLAY.md`                      |
 | Skills framework                              | `docs/frameworks/SKILLS.md`                             |
@@ -461,10 +461,18 @@ own dedicated branch, and you MUST confirm the base branch with the operator bef
    git fetch origin "$BASE_BRANCH"
    git worktree add ".claude/worktrees/${TASK##*/}" -b "$TASK" "origin/$BASE_BRANCH"
    cd ".claude/worktrees/${TASK##*/}"
-   # symlink node_modules from the main checkout to skip a per-worktree npm install:
+   # Reuse the main checkout's node_modules to skip a per-worktree npm install.
-   ln -s "$(git -C <main_checkout> rev-parse --show-toplevel)/node_modules" node_modules
+   # HARD LINKS (`cp -al`), never a symlink: ~5s for the whole tree and near-zero extra
+   # disk (the inodes are shared), and unlike a symlink it does not break the dev server.
+   cp -al "$(git -C <main_checkout> rev-parse --show-toplevel)/node_modules" node_modules
    ```
 
+   **Never `ln -s` node_modules.** Turbopack rejects a symlink that resolves outside the
+   project root, so `npm run dev` dies with a FATAL panic (`Symlink [project]/node_modules
+   is invalid, it points out of the filesystem root`) while typecheck, lint and the test
+   runners all keep passing — the error names "filesystem root", not the worktree, so it
+   reads like a Next/build bug and costs real time to trace (incident 2026-07-31, #9043).
+
    In Claude Code prefer the native `EnterWorktree` tool (it already creates worktrees under
    `.claude/worktrees/`): create the worktree with the command above, then call `EnterWorktree`
    with its `path`.

CONTRIBUTING.md

@@ -2,6 +2,11 @@
 
 Thank you for your interest in contributing! This guide covers everything you need to get started.
 
+For the official per-change workflow, start with the
+[Contribution Golden Path](docs/dev/CONTRIBUTION_GOLDEN_PATH.md). It maps provider, routing,
+UI/UX, i18n, CLI, database, and build/deploy changes to their contracts, focused tests, CI
+coverage, and reconciliation steps.
+
 ---
 
 ## Development Setup
@@ -198,10 +203,11 @@ Coverage notes:
 
 ### Pull Request Requirements
 
-Before opening a PR, run the focused loop for what you changed. The full unit suite
+Before opening a PR, use the
-(4 CI shards), Vitest, the **60%+** coverage gate, and the production build are CI's
+[Contribution Golden Path](docs/dev/CONTRIBUTION_GOLDEN_PATH.md) to run the focused loop for
-responsibility — running them locally adds no signal the PR checks will not already
+what you changed. The full unit suite (4 CI shards), Vitest, the **60%+** coverage gate, and
-give you, and on smaller machines it can saturate the host (#8084):
+the production build are CI's responsibility — running them locally adds no signal the PR
+checks will not already give you, and on smaller machines it can saturate the host (#8084):
 
 - Run the test files that cover your change: `node --import tsx/esm --test tests/unit/<file>.test.ts`
 - Run `npm run lint`
@@ -271,7 +277,7 @@ src/                        # TypeScript (.ts / .tsx)
 │   ├── a2a/                # Agent-to-Agent v0.3 protocol server
 │   ├── acp/                # Agent Communication Protocol registry
 │   ├── compliance/         # Compliance policy engine
-│   ├── db/                 # SQLite database layer (21 modules + 16 migrations)
+│   ├── db/                 # SQLite domain modules + 130 migrations
 │   ├── memory/             # Persistent conversational memory
 │   ├── oauth/              # OAuth providers, services, and utilities
 │   ├── skills/             # Extensible skill framework
@@ -281,7 +287,7 @@ src/                        # TypeScript (.ts / .tsx)
 ├── mitm/                   # MITM proxy (cert, DNS, target routing)
 ├── shared/
 │   ├── components/         # React components (.tsx)
-│   ├── constants/          # Provider definitions (177), MCP scopes, 14 routing strategies
+│   ├── constants/          # Provider definitions (290), MCP scopes, 19 routing strategies
 │   ├── utils/              # Circuit breaker, sanitizer, auth helpers
 │   └── validation/         # Zod v4 schemas
 └── sse/                    # SSE proxy pipeline
@@ -289,7 +295,7 @@ src/                        # TypeScript (.ts / .tsx)
 open-sse/                   # @omniroute/open-sse workspace
 ├── executors/              # 14 provider-specific request executors
 ├── handlers/               # 11 request handlers (chat, responses, embeddings, images, etc.)
-├── mcp-server/             # MCP server (25 tools, 3 transports, 10 scopes)
+├── mcp-server/             # MCP server (104 tools, 3 transports, 31 scopes)
 ├── services/               # 36+ services (combo, autoCombo, rateLimitManager, etc.)
 ├── translator/             # Format translators (OpenAI ↔ Claude ↔ Gemini ↔ Responses ↔ Ollama)
 ├── transformer/            # Responses API transformer

Dockerfile

@@ -236,6 +236,11 @@ FROM runner-base AS runner-cli
 # runner-base runs.
 USER root
 
+# The CLI image can use the internal ChatGPT Web (Codex) Chromium sidecar over
+# CDP without installing a second browser in this container.
+COPY --from=builder /app/node_modules/playwright-core ./node_modules/playwright-core
+COPY --from=builder /app/node_modules/playwright ./node_modules/playwright
+
 # Install system dependencies required by openclaw (git+ssh references).
 RUN --mount=type=cache,id=apt-cache,target=/var/cache/apt,sharing=locked \
   --mount=type=cache,id=apt-lists,target=/var/lib/apt/lists,sharing=locked \

README.md

@@ -241,12 +241,53 @@ curl http://localhost:20128/v1/chat/completions \
       <b>What Kimi's support powers:</b> Kimi's API credits power OmniRoute's AI-validated release pipeline — the <i>merge validation powered by Kimi K3</i> stage that reviews every pull request before it ships — plus day-to-day feature development. First-class Kimi support ships on both rails: the direct <a href="https://platform.kimi.ai?aff=omniroute">Kimi API</a> (<code>kimi-k3</code>) and the <a href="https://www.kimi.com/code?aff=omniroute">Kimi Code coding plan</a> (OAuth and API key). OmniRoute is also the first Brazilian open-source project in Kimi's support program. <a href="https://platform.kimi.ai?aff=omniroute"><b>Get a Kimi API key →</b></a>
     </td>
   </tr>
+  <tr>
+    <td align="center" width="150">
+      <a href="https://cheaperinference.com/?utm_source=omniroute">
+        <img src="public/providers/cheaperinference.svg" width="64" alt="Cheaper Inference"/>
+      </a>
+      <br/><b>Cheaper Inference</b><br/><sub>cheaperinference.com</sub><br/><br/>
+      <img src="https://img.shields.io/badge/Open_Source_Friend-31f889?style=flat-square&labelColor=04170d" alt="Open Source Friend"/>
+    </td>
+    <td>
+      Thanks to <b>Cheaper Inference</b>, an OmniRoute Open Source Friend, for backing this project! Cheaper Inference is a cost-ranked gateway that resells 42 frontier models — Claude, GPT-5.x, Gemini, Kimi K3, GLM, DeepSeek, Grok and MiniMax — behind one OpenAI-compatible endpoint, routing each request to the cheapest eligible provider without ever charging above the model maker's list price.
+      <br/><br/>
+      <b>First-class support in OmniRoute:</b> Chat Completions, the native <code>/v1/responses</code> endpoint, vision, tool calling and 3 image models (<code>grok-imagine</code>, <code>nano-banana-pro</code>, <code>nano-banana-2</code>, reachable as <code>cheaperinference/&lt;model&gt;</code>). <a href="https://cheaperinference.com/?utm_source=omniroute"><b>Get an API key →</b></a>
+    </td>
+  </tr>
 </table>
 
 <sub>Links tagged <code>aff=omniroute</code> are partner links. They fund the project at no extra cost to you.</sub>
 
 <br/>
 
+<details open>
+<summary><sub><b>🎟️ Affiliates Promo</b> — free signup coupons from providers we don't sponsor (click to expand)</sub></summary>
+
+<sub><i>This section is for referral/coupon codes only. Sponsored partnerships live in <b>🤝 Supported by our Open Source Friends</b> above. OmniRoute has no sponsorship or partnership with the providers listed here — these are public coupons anyone can use.</i></sub>
+
+<table>
+  <tr>
+    <td align="center" width="120">
+      <a href="https://agentrouter.org/register?aff=70LM">
+        <img src="public/providers/agentrouter.png" width="32" alt="AgentRouter"/>
+      </a>
+      <br/><sub><b>AgentRouter</b></sub><br/><sub>agentrouter.org</sub>
+    </td>
+    <td>
+      <sub><b><a href="https://agentrouter.org/register?aff=70LM">AgentRouter</a></b> — affiliate signup · <b>$100 free credits</b> on signup (free server, expect higher latency — best for testing, not production). First-class support in OmniRoute since <b>v3.8.50</b>: Chat Completions, the Anthropic-compatible wire format and the OpenAI-compatible path. Available models include <code>claude-opus-4-8</code>, <code>claude-opus-5</code>, <code>gpt-5.6-sol</code> and more. <b><a href="https://agentrouter.org/register?aff=70LM">Grab your $100 →</a></b></sub>
+      <br/><br/>
+      <sub>⚠️ <i>Affiliate link — OmniRoute has no sponsorship or partnership with this provider.</i></sub>
+    </td>
+  </tr>
+</table>
+
+<sub>Know another provider with a generous free signup coupon that benefits OmniRoute users? Open an issue and we'll add it here.</sub>
+
+</details>
+
+<br/>
+
 <div align="center">
 
 ## 🎯 Combos — The Flagship

THIRD_PARTY_NOTICES.md

@@ -0,0 +1,26 @@
+# Third-Party Notices
+
+## codex-chatgpt-web
+
+Parts of `open-sse/vendor/codex-chatgpt-web/` are adapted from
+[`miuuyy/codex-chatgpt-web`](https://github.com/miuuyy/codex-chatgpt-web), commit
+`55592fca0ba19a27f1b769cec8fff61ff340a785`.
+
+MIT License
+
+Copyright (c) 2026 codex-chatgpt-web contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of this software and
+associated documentation files (the "Software"), to deal in the Software without restriction,
+including without limitation the rights to use, copy, modify, merge, publish, distribute,
+sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all copies or substantial
+portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT
+NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT
+OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

bin/chatgpt-web-codex-mcp.mjs

@@ -0,0 +1,56 @@
+#!/usr/bin/env node
+
+import { existsSync } from "node:fs";
+import { dirname, join } from "node:path";
+import { fileURLToPath, pathToFileURL } from "node:url";
+
+const here = dirname(fileURLToPath(import.meta.url));
+const root = join(here, "..");
+
+export function resolveChatGptWebCodexMcpEntry(rootDir = root, exists = existsSync) {
+  const candidates = [
+    join(
+      rootDir,
+      "dist",
+      "open-sse",
+      "vendor",
+      "codex-chatgpt-web",
+      "adapters",
+      "chatgpt-web",
+      "mcp-server.js"
+    ),
+    join(
+      rootDir,
+      "open-sse",
+      "vendor",
+      "codex-chatgpt-web",
+      "adapters",
+      "chatgpt-web",
+      "mcp-server.ts"
+    ),
+  ];
+  return candidates.find((candidate) => exists(candidate)) ?? null;
+}
+
+export async function startChatGptWebCodexMcp(args = process.argv.slice(2), rootDir = root) {
+  const socketIndex = args.indexOf("--broker-socket");
+  const brokerSocketPath = socketIndex >= 0 ? args[socketIndex + 1] : undefined;
+  if (!brokerSocketPath) throw new Error("--broker-socket is required");
+  const entry = resolveChatGptWebCodexMcpEntry(rootDir);
+  if (!entry) throw new Error("ChatGPT Web (Codex) MCP entrypoint was not found");
+  if (entry.endsWith(".ts")) {
+    const { register } = await import("node:module");
+    register("tsx/esm", pathToFileURL(`${rootDir}/`));
+  }
+  const module = await import(pathToFileURL(entry).href);
+  await module.runChatGptMcpServer({ brokerSocketPath });
+}
+
+if (process.argv[1] && fileURLToPath(import.meta.url) === process.argv[1]) {
+  startChatGptWebCodexMcp().catch((error) => {
+    console.error(
+      `ChatGPT Web (Codex) MCP konnte nicht gestartet werden: ${error?.message || error}`
+    );
+    process.exit(1);
+  });
+}

changelog.d/fixes/8909-vertex-claude-rawpredict-streaming.md

@@ -0,0 +1 @@
+- **fix(executors):** Vertex AI now routes Claude models through the native Anthropic `rawPredict` endpoint instead of the generic OpenAI-compatible partner endpoint, and synthesizes a real streaming response so Claude-via-Vertex works with `stream: true` ([#8909](https://github.com/diegosouzapw/OmniRoute/pull/8909)) — thanks @wgordon17

changelog.d/fixes/9006-vertex-claude-catalog-dispatch.md

@@ -0,0 +1,14 @@
+- **fix(sse):** Claude reasoning-effort suffix ids (`-high`/`-low`/`-medium`/`-xhigh`) now strip
+  correctly on any provider serving a real Claude model, not just the direct Anthropic provider
+  ([#9006](https://github.com/diegosouzapw/OmniRoute/pull/9006))
+- **fix(sse):** the no-thinking (`no-think/`) catalog variant's provider-qualification bug — which
+  made it unusable outside the direct provider, both in the discovery catalog and the dashboard
+  playground — is fixed ([#9006](https://github.com/diegosouzapw/OmniRoute/pull/9006))
+- **fix(sse):** a single unrecognized model id on a Vertex connection no longer cools down every
+  other model on that connection for 2 minutes — Vertex 404s are now scoped to a per-model
+  lockout via `passthroughModels` instead of a connection-wide cooldown
+  ([#9006](https://github.com/diegosouzapw/OmniRoute/pull/9006))
+- **fix(sse):** Vertex `PERMISSION_DENIED` 403s are now disambiguated using Google's own
+  documented error format — a genuinely connection-wide cause (API disabled, project-level IAM
+  denial) still cools the whole connection, while a model-specific denial locks out only that
+  model ([#9006](https://github.com/diegosouzapw/OmniRoute/pull/9006))

changelog.d/fixes/9013-model-param-filter-save.md

@@ -0,0 +1 @@
+- **fix(dashboard):** model-level allowed/blocked param edits now persist when the compatibility popover is closed by clicking outside, and a failed save no longer clears the edit or reports success ([#9013](https://github.com/diegosouzapw/OmniRoute/pull/9013))

compression-core/.gitignore

@@ -0,0 +1,2 @@
+/target
+Cargo.lock

compression-core/Cargo.toml

@@ -0,0 +1,27 @@
+[workspace]
+resolver = "2"
+members = [
+    "crates/core-api",
+    "crates/tokenizer",
+    "crates/tests",
+    "crates/bench",
+    "crates/ffi",
+]
+
+[workspace.package]
+version = "0.1.0"
+edition = "2021"
+license = "MIT"
+repository = "https://github.com/Egorich-print/OmniRoute"
+
+[workspace.dependencies]
+core-api = { path = "crates/core-api" }
+tokenizer = { path = "crates/tokenizer" }
+tiktoken-rs = "0.6"
+serde = { version = "1", features = ["derive"] }
+serde_json = "1"
+thiserror = "2"
+
+[profile.release]
+lto = true
+codegen-units = 1

compression-core/README.md

@@ -0,0 +1,61 @@
+# compression-core
+
+Standalone Rust core for AI context optimization — tokenization, compression,
+hashing, translation primitives. Independent OSS library usable by OmniRoute,
+OpenCode, Cline, Roo, and any AI proxy. No OmniRoute imports anywhere.
+
+## Layout
+
+```
+compression-core/
+├── Cargo.toml            # workspace
+├── crates/
+│   ├── core-api/         # stable public API (traits + types) — no host deps
+│   ├── tokenizer/        # tiktoken (cl100k_base, o200k_base) — PORTED
+│   ├── tests/            # golden tests against fixtures/expected/
+│   ├── bench/            # criterion benchmarks
+│   └── ffi/              # N-API adapter (integration phase)
+├── fixtures/
+│   ├── tokenizer/        # JS-generated token counts (13 samples)
+│   └── expected/         # manifests
+└── scripts/
+    ├── generate-fixtures.ts  # JS reference output (source of truth)
+    └── verify-golden.ts      # regen + cargo test
+```
+
+## Porting order (per design)
+
+1. tiktoken (done — golden 100%)
+2. ionizer
+3. headroom
+4. caveman
+5. RTK (last — biggest, requires proven harness)
+
+## Golden pipeline
+
+```text
+fixtures → JS implementation → expected.json → Rust → assert_eq!
+```
+
+`node scripts/verify-golden.ts` regenerates fixtures from the current JS code
+and runs `cargo test -p compression-tests`. Until 100% match, JS stays in prod.
+
+## Measured baseline
+
+| Impl | Input | Cost |
+|---|---|---|
+| JS js-tiktoken (cl100k) | 230K chars | 37.9 ms |
+| Rust tiktoken-rs (cl100k) | ~440K chars | 21.4 ms |
+
+Per-char Rust is ~3x faster; golden output is byte-identical on all fixtures.
+
+## Status
+
+- [x] workspace + stable API (`core-api`)
+- [x] tokenizer port + golden tests (100% match)
+- [x] bench harness (criterion)
+- [ ] ionizer
+- [ ] headroom
+- [ ] caveman
+- [ ] RTK
+- [ ] N-API adapter

compression-core/crates/bench/Cargo.toml

@@ -0,0 +1,17 @@
+[package]
+name = "compression-bench"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+publish = false
+
+[dependencies]
+core-api = { workspace = true }
+tokenizer = { workspace = true }
+
+[dev-dependencies]
+criterion = "0.5"
+
+[[bench]]
+name = "tokenizer"
+harness = false

compression-core/crates/bench/benches/tokenizer.rs

@@ -0,0 +1,19 @@
+//! Criterion bench for the tokenizer. Baseline target: < 5 ms per 57K tokens
+//! (JS js-tiktoken measures ~38 ms on the same input).
+
+use core_api::TokenCounter;
+use criterion::{criterion_group, criterion_main, Criterion};
+use tokenizer::TiktokenCounter;
+
+fn bench_tokenizer(c: &mut Criterion) {
+    let counter = TiktokenCounter::default();
+    // ~230K chars ≈ 57K cl100k tokens (mirrors the measured JS baseline).
+    let text = "Hello world! This is a test of tokenization performance. \
+                The quick brown fox jumps over the lazy dog. "
+        .repeat(4000);
+
+    c.bench_function("cl100k_57k_tokens", |b| b.iter(|| counter.count(&text)));
+}
+
+criterion_group!(benches, bench_tokenizer);
+criterion_main!(benches);

compression-core/crates/core-api/Cargo.toml

@@ -0,0 +1,10 @@
+[package]
+name = "core-api"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+serde = { workspace = true }
+serde_json = { workspace = true }
+thiserror = { workspace = true }

compression-core/crates/core-api/src/lib.rs

@@ -0,0 +1,76 @@
+//! Stable public API of the compression core.
+//!
+//! This crate is intentionally free of any OmniRoute-specific types.
+//! It defines the contracts that every adapter (N-API, sidecar, CLI)
+//! implements, so algorithms stay independent of the host project.
+
+use serde::{Deserialize, Serialize};
+
+/// Role of a message in a conversation.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(rename_all = "lowercase")]
+pub enum Role {
+    System,
+    User,
+    Assistant,
+    Tool,
+}
+
+/// One chat message. Field-compatible with OpenAI `messages[]` entries.
+#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
+pub struct Message {
+    pub role: Role,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub content: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub name: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub tool_call_id: Option<String>,
+}
+
+/// Tokenizer encodings supported by the core.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+pub enum Encoding {
+    #[serde(rename = "cl100k_base")]
+    Cl100kBase,
+    #[serde(rename = "o200k_base")]
+    O200kBase,
+}
+
+/// Configuration for a compression pass.
+#[derive(Debug, Clone, PartialEq, Serialize, Deserialize, Default)]
+pub struct CompressionConfig {
+    /// Target token budget for the compressed messages.
+    pub budget_tokens: Option<u64>,
+    /// Engine stack priority hint (rtk=10, ionizer=13, headroom=15, ...).
+    pub stack_priority: Option<u32>,
+}
+
+/// Outcome of a compression pass.
+#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
+pub struct CompressionResult {
+    pub messages: Vec<Message>,
+    pub compressed: bool,
+    pub stats: Option<CompressionStats>,
+}
+
+#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
+pub struct CompressionStats {
+    pub saved_tokens: Option<u64>,
+    pub input_tokens: Option<u64>,
+    pub output_tokens: Option<u64>,
+}
+
+/// A token counter. Pure, stateless, thread-safe.
+pub trait TokenCounter {
+    fn count(&self, text: &str) -> usize;
+}
+
+/// A compressor. Pure, deterministic, stateless per call.
+pub trait Compressor {
+    fn compress(
+        &self,
+        messages: &[Message],
+        config: &CompressionConfig,
+    ) -> CompressionResult;
+}

compression-core/crates/ffi/Cargo.toml

@@ -0,0 +1,13 @@
+[package]
+name = "compression-ffi"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+publish = false
+
+[dependencies]
+core-api = { workspace = true }
+tokenizer = { workspace = true }
+
+# napi-rs bindings are added in the integration phase. This crate exists to
+# keep the N-API adapter out of the algorithm crates.

compression-core/crates/ffi/src/lib.rs

@@ -0,0 +1,8 @@
+//! N-API binding crate (integration phase).
+//!
+//! This crate is intentionally empty until the N-API phase. It will expose
+//! `count_tokens` / `compress` over napi-rs using the core-api traits, so the
+//! algorithms in `tokenizer` and the future `compression` crates stay free of
+//! any Node bindings.
+
+pub use core_api;

compression-core/crates/tests/Cargo.toml

@@ -0,0 +1,11 @@
+[package]
+name = "compression-tests"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+publish = false
+
+[dependencies]
+core-api = { workspace = true }
+tokenizer = { workspace = true }
+serde_json = { workspace = true }

compression-core/crates/tests/src/lib.rs

@@ -0,0 +1,66 @@
+//! Golden tests: run the Rust implementations against fixtures and compare
+//! byte-for-byte with the JS-produced `expected/` files.
+//!
+//! The `verify-golden.ts` script regenerates fixtures from the OmniRoute JS
+//! implementation. Until this crate passes 100% of golden fixtures, the JS
+//! implementation must NOT be replaced in production.
+
+use core_api::{Encoding, TokenCounter};
+use std::path::Path;
+use tokenizer::TiktokenCounter;
+
+const FIXTURES_DIR: &str = concat!(env!("CARGO_MANIFEST_DIR"), "/../../fixtures");
+
+fn fixture_path(relative: &str) -> String {
+    Path::new(FIXTURES_DIR).join(relative).to_string_lossy().into_owned()
+}
+
+#[test]
+fn tokenizer_golden_cl100k() {
+    let counter = TiktokenCounter::default();
+    let dir = fixture_path("tokenizer");
+    let entries = std::fs::read_dir(&dir).expect("fixtures/tokenizer must exist");
+    let mut checked = 0;
+    for entry in entries {
+        let path = entry.unwrap().path();
+        if path.extension().map(|e| e == "json").unwrap_or(false) {
+            let input: serde_json::Value =
+                serde_json::from_str(&std::fs::read_to_string(&path).unwrap()).unwrap();
+            let text = input["text"].as_str().unwrap();
+            let expected = input["cl100k_tokens"].as_u64().unwrap() as usize;
+            assert_eq!(
+                counter.count(text),
+                expected,
+                "cl100k mismatch on {}",
+                path.display()
+            );
+            checked += 1;
+        }
+    }
+    assert!(checked > 0, "no tokenizer fixtures found");
+}
+
+#[test]
+fn tokenizer_golden_o200k() {
+    let counter = TiktokenCounter::default();
+    let dir = fixture_path("tokenizer");
+    let entries = std::fs::read_dir(&dir).expect("fixtures/tokenizer must exist");
+    let mut checked = 0;
+    for entry in entries {
+        let path = entry.unwrap().path();
+        if path.extension().map(|e| e == "json").unwrap_or(false) {
+            let input: serde_json::Value =
+                serde_json::from_str(&std::fs::read_to_string(&path).unwrap()).unwrap();
+            let text = input["text"].as_str().unwrap();
+            let expected = input["o200k_tokens"].as_u64().unwrap() as usize;
+            assert_eq!(
+                counter.count_with_encoding(text, Encoding::O200kBase),
+                expected,
+                "o200k mismatch on {}",
+                path.display()
+            );
+            checked += 1;
+        }
+    }
+    assert!(checked > 0, "no tokenizer fixtures found");
+}

compression-core/crates/tokenizer/Cargo.toml

@@ -0,0 +1,13 @@
+[package]
+name = "tokenizer"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+core-api = { workspace = true }
+tiktoken-rs = { workspace = true }
+anyhow = "1"
+
+[dev-dependencies]
+serde_json = { workspace = true }

compression-core/crates/tokenizer/src/lib.rs

@@ -0,0 +1,71 @@
+//! Tiktoken token counter backed by `tiktoken-rs`.
+//!
+//! Port target: `src/shared/utils/tiktokenCounter.ts` in OmniRoute.
+//! Encodings: cl100k_base (default), o200k_base (Codex).
+
+use core_api::{Encoding, TokenCounter};
+use tiktoken_rs::tokenizer::Tokenizer;
+
+pub struct TiktokenCounter {
+    cl100k: tiktoken_rs::CoreBPE,
+    o200k: tiktoken_rs::CoreBPE,
+}
+
+impl TiktokenCounter {
+    pub fn new() -> Result<Self, anyhow::Error> {
+        let cl100k = tiktoken_rs::get_bpe_from_tokenizer(Tokenizer::Cl100kBase)?;
+        let o200k = tiktoken_rs::get_bpe_from_tokenizer(Tokenizer::O200kBase)?;
+        Ok(Self { cl100k, o200k })
+    }
+
+    pub fn count_with_encoding(&self, text: &str, encoding: Encoding) -> usize {
+        let bpe = match encoding {
+            Encoding::Cl100kBase => &self.cl100k,
+            Encoding::O200kBase => &self.o200k,
+        };
+        // CoreBPE::encode_with_special_tokens requires allocation; the
+        // plain encode is the closest equivalent to the JS byte-pair count.
+        bpe.encode_ordinary(text).len()
+    }
+}
+
+impl Default for TiktokenCounter {
+    fn default() -> Self {
+        Self::new().expect("tiktoken rank tables must load")
+    }
+}
+
+impl TokenCounter for TiktokenCounter {
+    fn count(&self, text: &str) -> usize {
+        self.count_with_encoding(text, Encoding::Cl100kBase)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn counts_known_tokens_cl100k() {
+        let counter = TiktokenCounter::default();
+        // "Hello world" is 2 tokens in cl100k_base.
+        assert_eq!(counter.count("Hello world"), 2);
+    }
+
+    #[test]
+    fn empty_string_is_zero() {
+        let counter = TiktokenCounter::default();
+        assert_eq!(counter.count(""), 0);
+    }
+
+    #[test]
+    fn o200k_differs_from_cl100k_on_emoji() {
+        let counter = TiktokenCounter::default();
+        let emoji = "🎉";
+        let cl100k = counter.count_with_encoding(emoji, Encoding::Cl100kBase);
+        let o200k = counter.count_with_encoding(emoji, Encoding::O200kBase);
+        // o200k has dedicated emoji tokens; counts may differ. Just assert both are > 0.
+        assert!(cl100k > 0);
+        assert!(o200k > 0);
+    }
+}

compression-core/fixtures/expected/tokenizer-manifest.json

@@ -0,0 +1,80 @@
+[
+  {
+    "id": "000",
+    "chars": 28,
+    "cl100k_tokens": 8,
+    "o200k_tokens": 8
+  },
+  {
+    "id": "001",
+    "chars": 44,
+    "cl100k_tokens": 10,
+    "o200k_tokens": 10
+  },
+  {
+    "id": "002",
+    "chars": 40,
+    "cl100k_tokens": 15,
+    "o200k_tokens": 12
+  },
+  {
+    "id": "003",
+    "chars": 38,
+    "cl100k_tokens": 15,
+    "o200k_tokens": 15
+  },
+  {
+    "id": "004",
+    "chars": 47,
+    "cl100k_tokens": 15,
+    "o200k_tokens": 15
+  },
+  {
+    "id": "005",
+    "chars": 100,
+    "cl100k_tokens": 100,
+    "o200k_tokens": 50
+  },
+  {
+    "id": "006",
+    "chars": 100,
+    "cl100k_tokens": 41,
+    "o200k_tokens": 23
+  },
+  {
+    "id": "007",
+    "chars": 10000,
+    "cl100k_tokens": 1250,
+    "o200k_tokens": 1250
+  },
+  {
+    "id": "008",
+    "chars": 1,
+    "cl100k_tokens": 1,
+    "o200k_tokens": 1
+  },
+  {
+    "id": "009",
+    "chars": 0,
+    "cl100k_tokens": 0,
+    "o200k_tokens": 0
+  },
+  {
+    "id": "010",
+    "chars": 49,
+    "cl100k_tokens": 18,
+    "o200k_tokens": 14
+  },
+  {
+    "id": "011",
+    "chars": 69,
+    "cl100k_tokens": 24,
+    "o200k_tokens": 24
+  },
+  {
+    "id": "012",
+    "chars": 405000,
+    "cl100k_tokens": 90001,
+    "o200k_tokens": 90001
+  }
+]

compression-core/fixtures/tokenizer/sample-000.json

@@ -0,0 +1,6 @@
+{
+  "id": "000",
+  "text": "Hello world! This is a test.",
+  "cl100k_tokens": 8,
+  "o200k_tokens": 8
+}

compression-core/fixtures/tokenizer/sample-001.json

@@ -0,0 +1,6 @@
+{
+  "id": "001",
+  "text": "The quick brown fox jumps over the lazy dog.",
+  "cl100k_tokens": 10,
+  "o200k_tokens": 10
+}

compression-core/fixtures/tokenizer/sample-002.json

@@ -0,0 +1,6 @@
+{
+  "id": "002",
+  "text": "🎉🎊 party time! emoji heavy sentence 🚀",
+  "cl100k_tokens": 15,
+  "o200k_tokens": 12
+}

compression-core/fixtures/tokenizer/sample-003.json

@@ -0,0 +1,6 @@
+{
+  "id": "003",
+  "text": "JSON:\n{\"name\":\"test\",\"values\":[1,2,3]}",
+  "cl100k_tokens": 15,
+  "o200k_tokens": 15
+}

compression-core/fixtures/tokenizer/sample-004.json

@@ -0,0 +1,6 @@
+{
+  "id": "004",
+  "text": "Code:\n```rust\nfn main() { println!(\"hi\"); }\n```",
+  "cl100k_tokens": 15,
+  "o200k_tokens": 15
+}

compression-core/fixtures/tokenizer/sample-005.json

@@ -0,0 +1,6 @@
+{
+  "id": "005",
+  "text": "😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀😀",
+  "cl100k_tokens": 100,
+  "o200k_tokens": 50
+}

compression-core/fixtures/tokenizer/sample-006.json

@@ -0,0 +1,6 @@
+{
+  "id": "006",
+  "text": "Поддерживается ли русский текст корректно? Проверяем длинное предложение с кириллицей и пунктуацией!",
+  "cl100k_tokens": 41,
+  "o200k_tokens": 23
+}

compression-core/fixtures/tokenizer/sample-008.json

@@ -0,0 +1,6 @@
+{
+  "id": "008",
+  "text": "t",
+  "cl100k_tokens": 1,
+  "o200k_tokens": 1
+}

compression-core/fixtures/tokenizer/sample-009.json

@@ -0,0 +1,6 @@
+{
+  "id": "009",
+  "text": "",
+  "cl100k_tokens": 0,
+  "o200k_tokens": 0
+}

compression-core/fixtures/tokenizer/sample-010.json

@@ -0,0 +1,6 @@
+{
+  "id": "010",
+  "text": "Mixed 🎯 unicode 中文 한국어 + english + numbers 12345",
+  "cl100k_tokens": 18,
+  "o200k_tokens": 14
+}

compression-core/fixtures/tokenizer/sample-011.json

@@ -0,0 +1,6 @@
+{
+  "id": "011",
+  "text": "function foo(a,b){return a+b*2;}\n\nconst x = foo(1,2);\nconsole.log(x);",
+  "cl100k_tokens": 24,
+  "o200k_tokens": 24
+}

compression-core/scripts/generate-fixtures.ts

@@ -0,0 +1,77 @@
+#!/usr/bin/env node
+/**
+ * Generates golden fixtures for compression-core from the OmniRoute JS
+ * implementation. Every fixture records: input text + expected token counts
+ * (cl100k / o200k) computed by the JS tokenizer.
+ *
+ * Usage:  node --import tsx/esm scripts/generate-fixtures.ts
+ * Output: fixtures/tokenizer/*.json, fixtures/conversations/*.json
+ *
+ * The Rust side (crates/tests) reads these and asserts equality. Until 100%
+ * of fixtures pass, the JS implementation must not be replaced.
+ */
+import { countTextTokens } from "../../src/shared/utils/tiktokenCounter.ts";
+import { mkdirSync, writeFileSync } from "node:fs";
+import { dirname, join } from "node:path";
+import { fileURLToPath } from "node:url";
+
+const HERE = dirname(fileURLToPath(import.meta.url));
+const ROOT = join(HERE, "..");
+const TOKENIZER_DIR = join(ROOT, "fixtures", "tokenizer");
+const EXPECTED_DIR = join(ROOT, "fixtures", "expected");
+
+mkdirSync(TOKENIZER_DIR, { recursive: true });
+mkdirSync(EXPECTED_DIR, { recursive: true });
+
+const SAMPLES = [
+  "Hello world! This is a test.",
+  "The quick brown fox jumps over the lazy dog.",
+  "🎉🎊 party time! emoji heavy sentence 🚀",
+  "JSON:\n{\"name\":\"test\",\"values\":[1,2,3]}",
+  "Code:\n```rust\nfn main() { println!(\"hi\"); }\n```",
+  "😀".repeat(50),
+  "Поддерживается ли русский текст корректно? Проверяем длинное предложение с кириллицей и пунктуацией!",
+  "a".repeat(10000),
+  "t".repeat(1),
+  "",
+  "Mixed 🎯 unicode 中文 한국어 + english + numbers 12345",
+  "function foo(a,b){return a+b*2;}\n\nconst x = foo(1,2);\nconsole.log(x);",
+];
+
+// A longer realistic conversation-style text (~230K chars) to mirror the
+// measured baseline and to stress the counter on large inputs.
+const LONG = ("The quick brown fox jumps over the lazy dog. ").repeat(9000);
+SAMPLES.push(LONG);
+
+const cl100k = (t) => countTextTokens(t);
+const o200k = (t) => countTextTokens(t, { provider: "codex", model: "codex/gpt-5.5" });
+
+let count = 0;
+for (const [idx, text] of SAMPLES.entries()) {
+  const id = String(idx).padStart(3, "0");
+  const record = {
+    id,
+    text,
+    cl100k_tokens: cl100k(text),
+    o200k_tokens: o200k(text),
+  };
+  writeFileSync(join(TOKENIZER_DIR, `sample-${id}.json`), JSON.stringify(record, null, 2));
+  count++;
+}
+
+// Also emit a combined manifest for quick scanning.
+writeFileSync(
+  join(EXPECTED_DIR, "tokenizer-manifest.json"),
+  JSON.stringify(
+    SAMPLES.map((t, idx) => ({
+      id: String(idx).padStart(3, "0"),
+      chars: t.length,
+      cl100k_tokens: cl100k(t),
+      o200k_tokens: o200k(t),
+    })),
+    null,
+    2
+  )
+);
+
+console.log(`Generated ${count} tokenizer fixtures + manifest in fixtures/`);

compression-core/scripts/verify-golden.ts

@@ -0,0 +1,45 @@
+#!/usr/bin/env node
+/**
+ * Verifies golden equivalence between the JS implementation and the Rust
+ * implementation.
+ *
+ * Rust side: runs `cargo test -p compression-tests` which asserts byte-level
+ * equality against fixtures/expected/. This script:
+ *   1. regenerates fixtures from the current JS implementation
+ *   2. runs cargo tests
+ *   3. reports pass/fail per fixture family
+ *
+ * Usage: node scripts/verify-golden.ts [--skip-generate]
+ */
+import { spawnSync } from "node:child_process";
+import { fileURLToPath } from "node:url";
+import { dirname, join } from "node:path";
+
+const HERE = dirname(fileURLToPath(import.meta.url));
+const CORE_DIR = join(HERE, "..");
+
+const skipGenerate = process.argv.includes("--skip-generate");
+
+if (!skipGenerate) {
+  console.log("[verify-golden] regenerating fixtures from JS implementation...");
+  const gen = spawnSync("node", ["--import", "tsx/esm", "scripts/generate-fixtures.ts"], {
+    cwd: CORE_DIR,
+    stdio: "inherit",
+  });
+  if (gen.status !== 0) {
+    console.error("FAIL: fixture generation exited with", gen.status);
+    process.exit(1);
+  }
+}
+
+console.log("[verify-golden] running Rust golden tests...");
+const run = spawnSync("cargo", ["test", "-p", "compression-tests"], {
+  cwd: CORE_DIR,
+  stdio: "inherit",
+});
+if (run.status !== 0) {
+  console.error("FAIL: Rust golden tests exited with", run.status);
+  process.exit(1);
+}
+
+console.log("[verify-golden] ALL GOLDEN TESTS PASSED ✅");

config/quality/dashboard-typecheck-baseline.json

@@ -127,12 +127,6 @@
   "src/app/(dashboard)/dashboard/providers/[id]/components/ConnectionsListPanel.tsx": {
     "TS2322": 2
   },
-  "src/app/(dashboard)/dashboard/providers/[id]/components/CustomModelsSection.tsx": {
-    "TS2739": 1
-  },
-  "src/app/(dashboard)/dashboard/providers/[id]/components/ModelCompatPopover.tsx": {
-    "TS2304": 5
-  },
   "src/app/(dashboard)/dashboard/providers/[id]/components/ProviderModalsPanel.tsx": {
     "TS2322": 3,
     "TS2739": 1,
@@ -147,9 +141,6 @@
   "src/app/(dashboard)/dashboard/providers/[id]/components/ProviderPlaygroundPanel.tsx": {
     "TS2503": 1
   },
-  "src/app/(dashboard)/dashboard/providers/[id]/components/__tests__/phase1d.test.tsx": {
-    "TS2739": 2
-  },
   "src/app/(dashboard)/dashboard/providers/[id]/components/modals/EditConnectionModal.tsx": {
     "TS2322": 1
   },

config/quality/dependency-allowlist.json

@@ -121,6 +121,8 @@
     "tailwind-merge",
     "tailwindcss",
     "tls-client-node",
+    "turndown",
+    "turndown-plugin-gfm",
     "tsup",
     "tsx",
     "type-coverage",

config/quality/file-size-baseline.json

@@ -1,4 +1,5 @@
 {
+  "_rebaseline_2026_07_30_9006_vertex_claude_catalog_dispatch": "PR #9006 (fix/vertex-claude-catalog-dispatch): three files, two causes. (1) src/sse/handlers/chat.ts 1845->1846 (+1): NOT this PR's own growth — this PR never touches chat.ts at all. Measured 1846 (split(\"\\n\").length) at this PR's own merge-base (before any of its 11 commits), so the drift was already inherited from already-merged PRs on release/v3.8.50 (fast-gates PR->release do not run check:file-size, same root cause as _rebaseline_2026_07_25_v3849_basered_filesize and _rebaseline_2026_07_02_5798_release_green) — no offending branch left to fix. (2) src/sse/services/auth.ts 2508->2512 (+4 net, after extraction — see below) and open-sse/handlers/chatCore.ts 5020->5023 (+3, comment-only): genuine own growth. auth.ts adds Vertex 403 PERMISSION_DENIED disambiguation (Google's google.rpc.ErrorInfo proto distinguishes a connection-wide cause — SERVICE_DISABLED, or IAM_PERMISSION_DENIED against a project-level resource — from a model-specific one scoped to a .../models/<id> resource), added mid-PR after a quality-gate reviewer flagged the plan's originally-accepted \"Vertex 403 always -> per-model lockout\" trade-off. The actual classification logic (~40 lines) was EXTRACTED into a new leaf module src/sse/services/vertexErrorClassifier.ts (mirrors the googApiKeyAuth.ts precedent, _rebaseline_2026_07_14_7034_goog_api_key), leaving only the irreducible call-site wiring in the frozen file: a 1-line import plus widening the existing #3027 per-model-403 guard condition. chatCore.ts's +3 is a pure comment expansion (no functional change) clarifying that the adjacent effort-suffix strip is no longer unconditional for every provider, requested by a separate quality-gate code-reviewer finding; not extractable (it's a comment). Auth.ts's disambiguation logic covered by 3 new test cases in tests/unit/vertex-passthrough-model-lockout.test.ts (SERVICE_DISABLED, IAM_PERMISSION_DENIED+model-resource, IAM_PERMISSION_DENIED+project-resource) plus a 4th regression test for a multi-detail-body correlation bug (reason and resource must be read from the SAME ErrorInfo detail, not independently regexed across the whole body) found by an adversarial quality-gate pass and fixed before merge.",
   "_rebaseline_2026_07_24_8470_hyperagent_sticky_thread": "PR #8470 (artickc, fix/hyperagent-tool-loop-thread-sticky) own growth: open-sse/executors/hyperagent.ts 936->1025 (wc -l; check-file-size.mjs counts via split(\"\\n\").length so the gate sees 937->1026, +89, crosses the 1000 cap). Fixes a real bug where a reverse-conversion proxy (text-Intent/JSON to Claude Code native tool_calls) rewrites assistant messages between agentic tool-loop turns, breaking HyperAgent's conversation-prefix fingerprint and cold-starting the thread mid tool-loop. Adds Anthropic tool_use/tool_result flattening to extractMessageText() plus a new rootUserFingerprint()/root-key lookup tier in resolveHyperAgentThreadBinding()/storeHyperAgentThreadAfterTurn() so the thread stays sticky across the tool loop. Cohesive additions inside the existing single-file executor; not extractable without splitting the executor mid-request-flow. Covered by tests/unit/executor-hyperagent.test.ts (19/19, +5 new cases for tool_result/tool_use flattening + root-key stickiness). Pre-merge review flagged a cross-conversation root-key collision risk (tracked in the PR's own mandatory pre-merge checklist, not yet addressed) — unrelated to this file-size ratchet, tracked separately by /fix-prs.",
   "_rebaseline_2026_07_25_8494_capability_filter_fail_closed": "PR #8494 (fix/capability-filters-fail-closed, #8488) own growth: open-sse/services/combo.ts 3640->3693 (+53) adds a fail-closed guard after filterTargetsByRequestCompatibility() — when every eligible target is excluded by request-capability filtering (vision/tools/etc) instead of quota/health, the combo now returns an explicit `capability_mismatch` 400 (describeCapabilityFilterExhaustion, imported from combo/comboStructure.ts) rather than silently falling through to a generic no-targets error, plus a `compatFilterFailOpen` escape hatch (combo config OR settings) mirrored at both the main/auto and round-robin call sites for symmetry. combo/comboStructure.ts (previously under cap, un-frozen) grows 794->918 (+124) — new home for describeCapabilityFilterExhaustion + providerSupportsEmulatedToolCalling (#5240 emulated tool-calling exemption so fail-closed does not regress prompt-emulation-only combos like all-chatgpt-web). Irreducible orchestration wiring at the existing filter chokepoint (same precedent as #7301's universal-cooldown-retry generalization). Companion test tests/unit/combo-routing-engine.test.ts 3409->3449 (+40, fail-closed/fail-open coverage across both call sites) also rebaselined. Covered by tests/unit/8488-capability-filter-fail-closed.test.ts (new) + 95/95 passing across both files. Structural shrink of combo.ts tracked in #3501.",
   "_rebaseline_2026_07_25_8499_ts7_result_union_predicates": "PR #8499 (backryun, chore/ts7-types-executor-scattered) own growth: muse-spark-web.ts 1396->1405 (+9, irreducible). Under this workspace's `strictNullChecks: false`, the boolean-literal discriminant on `GraphqlResult` (`{ ok: true } | { ok: false; error: string }`) narrows the positive `.ok===true` branch but leaves `!result.ok` at the full union under TS7, making `.error` unreachable to the checker at the two call sites (warmup, mode-switch). Fixed by adding a single `isGraphqlFailure()` type-predicate helper (doc comment + 3-line body) reused at both call sites instead of duplicating the predicate inline — not extractable to a shared module without splitting a single-file executor's local narrowing helper out of its own file. Covered by the existing muse-spark-web executor test suite (no behavior change, pure narrowing fix).",
@@ -177,7 +178,7 @@
     "tests/unit/account-fallback-service.test.ts": 1563,
     "tests/unit/batch_api.test.ts": 1324,
     "tests/unit/cc-compatible-provider.test.ts": 1217,
-    "tests/unit/chatcore-translation-paths.test.ts": 2769,
+    "tests/unit/chatcore-translation-paths.test.ts": 2776,
     "tests/unit/chatgpt-web.test.ts": 3148,
     "tests/unit/combo-routing-engine.test.ts": 3449,
     "tests/unit/db-migration-runner.test.ts": 1499,
@@ -342,14 +343,14 @@
     "_rebaseline_pr1043_minimax_tts": "Upstream port decolua/9router#1043 (toanalien) own growth: audioSpeech.ts 965->1061 (+96). Adds MiniMax T2A v2 TTS dispatch (handleMinimaxSpeech + hexToBytes helper) — provider entry was already in audioRegistry (format: minimax-tts) but no handler existed, falling through to the OpenAI-compatible default that fails (T2A has custom shape + hex-encoded audio + base_resp envelope). New branch sits next to the other inline provider branches (xiaomi-mimo, coqui, tortoise, aws-polly) — extracting would just create indirection. Covered by tests/unit/minimax-tts-1043.test.ts (3 tests, GREEN: success, base_resp error, invalid-hex).",
     "_rebaseline_pr4592_exclude_exhausted_auto": "Reconcile #4592 already-merged growth: combo.ts 2991->3036 (+45, terminal-status quota-cutoff exclusion in buildAutoCandidates + opt-in gate). Fast-gate PR->release does not run check:file-size.",
     "open-sse/executors/antigravity.ts": 1528,
-    "open-sse/executors/base.ts": 1562,
+    "open-sse/executors/base.ts": 1578,
     "open-sse/executors/chatgpt-web.ts": 3241,
     "open-sse/executors/codex.ts": 1534,
     "open-sse/executors/cursor.ts": 1560,
     "open-sse/executors/deepseek-web.ts": 1148,
     "open-sse/executors/grok-web.ts": 1044,
     "open-sse/executors/muse-spark-web.ts": 1405,
-    "open-sse/handlers/chatCore.ts": 5020,
+    "open-sse/handlers/chatCore.ts": 5023,
     "open-sse/handlers/imageGeneration.ts": 3101,
     "open-sse/handlers/responseSanitizer.ts": 1115,
     "open-sse/handlers/search.ts": 1536,
@@ -365,7 +366,7 @@
     "open-sse/services/rateLimitManager.ts": 1060,
     "open-sse/translator/response/openai-responses.ts": 1174,
     "open-sse/utils/cursorAgentProtobuf.ts": 1505,
-    "open-sse/utils/stream.ts": 2887,
+    "open-sse/utils/stream.ts": 2889,
     "src/app/(dashboard)/dashboard/HomePageClient.tsx": 1381,
     "src/app/(dashboard)/dashboard/analytics/ComboHealthTab.tsx": 1031,
     "src/app/(dashboard)/dashboard/api-manager/ApiManagerPageClient.tsx": 3117,
@@ -400,8 +401,8 @@
     "src/shared/components/RequestLoggerV2.tsx": 1629,
     "src/shared/components/analytics/charts.tsx": 1035,
     "src/shared/services/cliRuntime.ts": 1122,
-    "src/sse/handlers/chat.ts": 1845,
+    "src/sse/handlers/chat.ts": 1846,
-    "src/sse/services/auth.ts": 2508,
+    "src/sse/services/auth.ts": 2512,
     "tests/unit/account-fallback-service.test.ts": 1572,
     "tests/unit/provider-validation-specialty.test.ts": 2980,
     "open-sse/executors/hyperagent.ts": 1026
@@ -413,5 +414,6 @@
   "_rebaseline_2026_07_28_8860_tokenrefresh_projectid": "PR #8860 (fix/antigravity-projectid-centralized) own test growth: tests/unit/token-refresh-service.test.ts 1311->1378 (+67 = 4 cases covering projectId discovery on the tokenRefresh.ts path — the Dashboard/health-check refresh route, which #8842 did not reach since that fixed the executor path). Covered by the same file.",
   "_rebaseline_2026_07_28_8861_xiaomi_token_plan": "PR #8861 (feat/xiaomi-token-plan-protocol-selector) own growth: EditConnectionModal.tsx 1283->1316 (+33 = the per-connection API-protocol selector field) and open-sse/executors/base.ts 1540->1562 (+22 = alternate-format resolution at the existing buildUrl/headers chokepoint). Both are irreducible wiring at existing call sites.",
   "_rebaseline_2026_07_28_8863_firefly_detail_level": "PR #8863 (fix/adobe-firefly-gpt-detail-level-max) own growth: adobeFireflyClient.ts 2317->2322 (+5 = gpt-image detailLevel defaulting to maximal at the existing payload-build site). Covered by tests/unit/adobe-firefly.test.ts.",
-  "_rebaseline_2026_07_29_8281_home_quickstart_prefetch": "Release v3.8.49 base-red fix (no PR — captain sweep): src/app/(dashboard)/dashboard/HomePageClient.tsx 1377->1381 (+4). #8292 added prefetch={false} to the sidebar but left /home's five quick-start Links prefetching, so first paint still fired 12 speculative RSC requests — caught by navigation.spec.ts only after the e2e helper bug (APP_ROUTE_PATTERN missing /home) was repaired in the same cycle. Growth is the five prefetch attributes; it was offset first by extracting the repeated className literals (INLINE_LINK x4, DOCS_LINK x1), which collapsed five wrapped <Link> blocks back to one line each — a naive fix measured 1391. Guard: tests/unit/sidebar-prefetch-policy-8281.test.ts."
+  "_rebaseline_2026_07_29_8281_home_quickstart_prefetch": "Release v3.8.49 base-red fix (no PR — captain sweep): src/app/(dashboard)/dashboard/HomePageClient.tsx 1377->1381 (+4). #8292 added prefetch={false} to the sidebar but left /home's five quick-start Links prefetching, so first paint still fired 12 speculative RSC requests — caught by navigation.spec.ts only after the e2e helper bug (APP_ROUTE_PATTERN missing /home) was repaired in the same cycle. Growth is the five prefetch attributes; it was offset first by extracting the repeated className literals (INLINE_LINK x4, DOCS_LINK x1), which collapsed five wrapped <Link> blocks back to one line each — a naive fix measured 1391. Guard: tests/unit/sidebar-prefetch-policy-8281.test.ts.",
+  "_rebaseline_2026_08_02_v3850_agentrouter_responses": "Release v3.8.50 AgentRouter/Codex compatibility reconciliation. open-sse/executors/base.ts 1562->1578: #9190 wires AgentRouter's selected Claude/OpenAI/Responses protocol through the existing executor URL, auth, identity-header and fingerprint chokepoints; the reusable alternate resolver remains outside base.ts. open-sse/utils/stream.ts 2887->2889: #9213 evaluates Responses ID and usage normalization independently so response.completed always receives finite usage.total_tokens instead of short-circuiting after an ID rewrite. tests/unit/chatcore-translation-paths.test.ts 2769->2776: #9191 updates the existing Claude-Code bridge assertions for the dynamic AgentRouter wire image. PR #9224 offsets its own chatCore growth by extracting the AgentRouter protocol decisions into chatCore/agentRouterProtocol.ts, leaving chatCore below its frozen ceiling. Covered by agentrouter executor/chatCore protocol tests, chatcore translation-path tests, and responses-commentary-passthrough tests."
 }

docker-compose.prod.yml

@@ -46,6 +46,8 @@ services:
     depends_on:
       redis:
         condition: service_healthy
+      chatgpt-web-codex-browser:
+        condition: service_started
     build:
       context: .
       target: runner-cli
@@ -67,6 +69,7 @@ services:
       - HOSTNAME=0.0.0.0
       - DATA_DIR=/app/data
       - OMNIROUTE_BASE_PATH=${OMNIROUTE_BASE_PATH:-}
+      - CHATGPT_WEB_CODEX_CDP_URL=http://chatgpt-web-codex-browser:9223
     ports:
       - "${PROD_DASHBOARD_PORT:-20130}:${DASHBOARD_PORT:-${PORT:-20128}}"
       - "${PROD_API_PORT:-20131}:${API_PORT:-20129}"
@@ -80,7 +83,19 @@ services:
       retries: 3
       start_period: 15s
 
+  chatgpt-web-codex-browser:
+    build:
+      context: .
+      dockerfile: docker/chatgpt-web-codex-browser/Dockerfile
+    image: omniroute:chatgpt-web-codex-browser
+    restart: unless-stopped
+    shm_size: "2gb"
+    volumes:
+      - chatgpt-web-codex-browser-prod-data:/browser-profile
+
 volumes:
+  chatgpt-web-codex-browser-prod-data:
+    name: omniroute-chatgpt-web-codex-browser-prod-data
   omniroute-prod-data:
     name: omniroute-prod-data
   redis-prod-data:

docker-compose.yml

@@ -98,6 +98,21 @@ services:
       args:
         OMNIROUTE_BASE_PATH: ${OMNIROUTE_BASE_PATH:-}
     image: omniroute:web
+    depends_on:
+      chatgpt-web-codex-browser:
+        condition: service_started
+    environment:
+      - DATA_DIR=/app/data
+      - PORT=${PORT:-20128}
+      - DASHBOARD_PORT=${DASHBOARD_PORT:-20128}
+      - API_PORT=${API_PORT:-20129}
+      - API_HOST=${API_HOST:-0.0.0.0}
+      - LIVE_WS_PORT=${LIVE_WS_PORT:-20132}
+      - LIVE_WS_HOST=${LIVE_WS_HOST:-0.0.0.0}
+      - LIVE_WS_ALLOWED_ORIGINS=${LIVE_WS_ALLOWED_ORIGINS:-http://localhost:20128,http://127.0.0.1:20128}
+      - REDIS_URL=${REDIS_URL:-redis://redis:6379}
+      - OMNIROUTE_BASE_PATH=${OMNIROUTE_BASE_PATH:-}
+      - CHATGPT_WEB_CODEX_CDP_URL=http://chatgpt-web-codex-browser:9223
     ports:
       - "${DASHBOARD_PORT:-20128}:${DASHBOARD_PORT:-20128}"
       - "${API_PORT:-20129}:${API_PORT:-20129}"
@@ -105,6 +120,20 @@ services:
     profiles:
       - web
 
+  # Internal-only Chromium runtime for ChatGPT Web (Codex). No CDP or browser
+  # UI port is published to the host.
+  chatgpt-web-codex-browser:
+    build:
+      context: .
+      dockerfile: docker/chatgpt-web-codex-browser/Dockerfile
+    image: omniroute:chatgpt-web-codex-browser
+    restart: unless-stopped
+    shm_size: "2gb"
+    volumes:
+      - chatgpt-web-codex-browser-data:/browser-profile
+    profiles:
+      - web
+
   # ── Profile: cli (CLIs installed inside container) ─────────────────
   omniroute-cli:
     <<: *common
@@ -252,6 +281,8 @@ services:
       - cliproxyapi
 
 volumes:
+  chatgpt-web-codex-browser-data:
+    name: omniroute-chatgpt-web-codex-browser-data
   cliproxyapi-data:
     name: cliproxyapi-data
   redis-data:

docker/chatgpt-web-codex-browser/Dockerfile

@@ -0,0 +1,10 @@
+FROM mcr.microsoft.com/playwright:v1.62.0-noble
+
+USER root
+RUN mkdir -p /browser-profile && chown -R pwuser:pwuser /browser-profile
+COPY --chown=pwuser:pwuser docker/chatgpt-web-codex-browser/cdp-proxy.mjs /opt/cdp-proxy.mjs
+USER pwuser
+
+EXPOSE 9223
+
+CMD ["/bin/sh", "-lc", "node /opt/cdp-proxy.mjs & exec $(find /ms-playwright -path '*/chrome-linux/chrome' -type f | head -n 1) --headless=new --no-sandbox --disable-dev-shm-usage --remote-debugging-port=9222 --user-data-dir=/browser-profile about:blank"]

docker/chatgpt-web-codex-browser/cdp-proxy.mjs

@@ -0,0 +1,72 @@
+import http from "node:http";
+import net from "node:net";
+
+const listenPort = 9223;
+const upstreamHost = "127.0.0.1";
+const upstreamPort = 9222;
+
+function proxyHeaders(headers) {
+  const next = { ...headers, host: `${upstreamHost}:${upstreamPort}` };
+  delete next.connection;
+  delete next.upgrade;
+  return next;
+}
+
+const server = http.createServer((request, response) => {
+  const upstream = http.request(
+    {
+      host: upstreamHost,
+      port: upstreamPort,
+      method: request.method,
+      path: request.url,
+      headers: proxyHeaders(request.headers),
+    },
+    (upstreamResponse) => {
+      const chunks = [];
+      upstreamResponse.on("data", (chunk) => chunks.push(chunk));
+      upstreamResponse.on("end", () => {
+        let body = Buffer.concat(chunks);
+        const contentType = String(upstreamResponse.headers["content-type"] || "");
+        if (contentType.includes("application/json")) {
+          body = Buffer.from(
+            body
+              .toString("utf8")
+              .replaceAll(`ws://${upstreamHost}:${upstreamPort}`, `ws://${request.headers.host}`)
+          );
+        }
+        const headers = { ...upstreamResponse.headers, "content-length": String(body.length) };
+        response.writeHead(upstreamResponse.statusCode || 502, headers);
+        response.end(body);
+      });
+    }
+  );
+  upstream.on("error", () => {
+    response.writeHead(503, { "content-type": "application/json" });
+    response.end(JSON.stringify({ error: "CDP browser is starting" }));
+  });
+  request.pipe(upstream);
+});
+
+server.on("upgrade", (request, socket, head) => {
+  const upstream = net.connect(upstreamPort, upstreamHost, () => {
+    const upgradeHeaders = {
+      ...request.headers,
+      host: `${upstreamHost}:${upstreamPort}`,
+      connection: "Upgrade",
+      upgrade: "websocket",
+    };
+    const headers = Object.entries(upgradeHeaders)
+      .flatMap(([name, value]) =>
+        Array.isArray(value) ? value.map((item) => `${name}: ${item}`) : [`${name}: ${value}`]
+      )
+      .join("\r\n");
+    upstream.write(
+      `${request.method} ${request.url} HTTP/${request.httpVersion}\r\n${headers}\r\n\r\n`
+    );
+    if (head.length > 0) upstream.write(head);
+    socket.pipe(upstream).pipe(socket);
+  });
+  upstream.on("error", () => socket.destroy());
+});
+
+server.listen(listenPort, "0.0.0.0");

docker/devin-bridge/Dockerfile

@@ -0,0 +1,57 @@
+FROM node:26.0.0-bookworm-slim
+
+ARG CLAUDE_CODE_VERSION=2.1.220
+ARG DEVIN_CLI_VERSION=3000.2.17
+ARG TARGETARCH
+
+RUN apt-get update \
+  && apt-get install -y --no-install-recommends ca-certificates curl git bash python3 make g++ tini \
+  && rm -rf /var/lib/apt/lists/* \
+  && npm install --global "@anthropic-ai/claude-code@${CLAUDE_CODE_VERSION}"
+
+RUN set -eu; \
+  case "${TARGETARCH}" in \
+    amd64) devin_arch=x86_64-unknown-linux; devin_sha=f0e1e9363afc6ee68c4ef87bab4aeb7ff5cc08a5fa838350ef3ceefdbb2a2be2 ;; \
+    arm64) devin_arch=aarch64-unknown-linux; devin_sha=116dc71ef085a922bc3ff0ea0377d4b26c529a431d58246e36572913e2d25624 ;; \
+    *) echo "Unsupported TARGETARCH=${TARGETARCH}" >&2; exit 1 ;; \
+  esac; \
+  curl -fsSL "https://static.devin.ai/cli/${DEVIN_CLI_VERSION}/devin-${DEVIN_CLI_VERSION}-${devin_arch}.tar.gz" -o /tmp/devin.tar.gz; \
+  echo "${devin_sha}  /tmp/devin.tar.gz" | sha256sum -c -; \
+  tar -xzf /tmp/devin.tar.gz -C /tmp; \
+  install -m 0755 "$(find /tmp -type f -name devin | head -1)" /usr/local/bin/devin; \
+  rm -rf /tmp/devin.tar.gz /tmp/devin-*
+
+RUN groupadd --gid 10001 bridge \
+  && useradd --uid 10001 --gid bridge --create-home --home-dir /home/bridge --shell /bin/bash bridge \
+  && mkdir -p /opt/omniroute /workspace \
+  && chown -R bridge:bridge /opt/omniroute /workspace
+
+WORKDIR /opt/omniroute
+USER bridge
+COPY --chown=bridge:bridge package.json package-lock.json .npmrc ./
+RUN npm ci --ignore-scripts --no-audit --fund=false
+COPY --chown=bridge:bridge . .
+RUN npm rebuild better-sqlite3 || true
+
+ENV HOME=/home/bridge \
+  CLAUDE_CONFIG_DIR=/home/bridge/.claude-devin-isolated \
+  DEVIN_AGENTIC_HOME=/home/bridge \
+  DATA_DIR=/home/bridge/.omniroute-isolated \
+  SQLITE_FILE=/home/bridge/.omniroute-isolated/storage.sqlite \
+  CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 \
+  DISABLE_TELEMETRY=1 \
+  DISABLE_ERROR_REPORTING=1 \
+  DISABLE_AUTOUPDATER=1 \
+  CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1 \
+  NEXT_TELEMETRY_DISABLED=1
+
+RUN mkdir -p /home/bridge/.claude-devin-isolated /home/bridge/.local/share/devin \
+  /home/bridge/.omniroute-isolated
+
+RUN DATA_DIR=/tmp/omniroute-build-data \
+  SQLITE_FILE=/tmp/omniroute-build-data/storage.sqlite \
+  npm run build \
+  && rm -rf /tmp/omniroute-build-data
+
+ENTRYPOINT ["/usr/bin/tini", "--"]
+CMD ["bash"]

docker/devin-bridge/compose.yml

@@ -0,0 +1,218 @@
+name: omniroute-devin-bridge
+
+x-isolated-environment: &isolated-environment
+  HOME: /home/bridge
+  CLAUDE_CONFIG_DIR: /home/bridge/.claude-devin-isolated
+  DEVIN_AGENTIC_HOME: /home/bridge
+  DATA_DIR: /home/bridge/.omniroute-isolated
+  SQLITE_FILE: /home/bridge/.omniroute-isolated/storage.sqlite
+  ANTHROPIC_BASE_URL: http://omniroute:20128
+  ANTHROPIC_AUTH_TOKEN: sk-local-devin-gateway
+  CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC: "1"
+  DISABLE_TELEMETRY: "1"
+  DISABLE_ERROR_REPORTING: "1"
+  DISABLE_AUTOUPDATER: "1"
+  CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY: "1"
+  DEVIN_BRIDGE_MODEL: ${DEVIN_BRIDGE_MODEL:-devin-cli-agentic/swe-1-7}
+  ANTHROPIC_MODEL: ${DEVIN_BRIDGE_MODEL:-devin-cli-agentic/swe-1-7}
+  ANTHROPIC_DEFAULT_SONNET_MODEL: ${DEVIN_BRIDGE_SONNET_MODEL:-devin-cli-agentic/swe-1-7}
+  ANTHROPIC_DEFAULT_OPUS_MODEL: ${DEVIN_BRIDGE_OPUS_MODEL:-devin-cli-agentic/swe-1-7}
+  ANTHROPIC_DEFAULT_HAIKU_MODEL: ${DEVIN_BRIDGE_HAIKU_MODEL:-devin-cli-agentic/swe-1-7}
+  CLAUDE_CODE_SUBAGENT_MODEL: ${DEVIN_BRIDGE_SUBAGENT_MODEL:-devin-cli-agentic/swe-1-7}
+  REQUIRE_API_KEY: "true"
+  OMNIROUTE_API_KEY: sk-local-devin-gateway
+
+x-runtime: &runtime
+  image: omniroute-devin-bridge:local
+  build:
+    context: ../..
+    dockerfile: docker/devin-bridge/Dockerfile
+    args:
+      CLAUDE_CODE_VERSION: 2.1.220
+      DEVIN_CLI_VERSION: 3000.2.17
+  user: "10001:10001"
+  read_only: true
+  tmpfs:
+    - /tmp:rw,noexec,nosuid,nodev,size=256m
+    - /opt/omniroute/.source:rw,nosuid,nodev,size=16m,uid=10001,gid=10001
+  cap_drop: [ALL]
+  security_opt: [no-new-privileges:true]
+  environment: *isolated-environment
+  networks: [bridge-internal]
+
+services:
+  omniroute:
+    <<: *runtime
+    profiles: [offline]
+    hostname: omniroute
+    environment:
+      <<: *isolated-environment
+      CLI_DEVIN_AGENTIC_BIN: /opt/omniroute/docker/devin-bridge/mock-devin.mjs
+      DEVIN_BRIDGE_MOCK_LOG: /evidence/mock-acp.jsonl
+    command: ["npm", "run", "start"]
+    healthcheck:
+      test:
+        [
+          "CMD",
+          "node",
+          "-e",
+          "fetch('http://127.0.0.1:20128/healthz').then(r=>process.exit(r.ok?0:1)).catch(()=>process.exit(1))",
+        ]
+      interval: 2s
+      timeout: 2s
+      retries: 60
+    volumes:
+      - omniroute-offline-data:/home/bridge/.omniroute-isolated
+      - ../../.sandbox/evidence:/evidence
+      - ./mock-devin.mjs:/opt/omniroute/docker/devin-bridge/mock-devin.mjs:ro
+
+  claude:
+    <<: *runtime
+    profiles: [offline]
+    depends_on:
+      omniroute:
+        condition: service_healthy
+      claude-egress-guard:
+        condition: service_healthy
+    working_dir: /workspace
+    command: ["bash", "/opt/omniroute/docker/devin-bridge/run-claude-e2e.sh"]
+    environment:
+      <<: *isolated-environment
+      NODE_USE_ENV_PROXY: "1"
+      HTTP_PROXY: http://claude-egress-guard:8080
+      HTTPS_PROXY: http://claude-egress-guard:8080
+      NO_PROXY: omniroute
+    volumes:
+      - claude-isolated-config:/home/bridge/.claude-devin-isolated
+      - ../../.sandbox/e2e-workspace:/workspace
+      - ../../.sandbox/evidence:/evidence
+      - ./run-claude-e2e.sh:/opt/omniroute/docker/devin-bridge/run-claude-e2e.sh:ro
+
+  contract:
+    <<: *runtime
+    profiles: [offline]
+    depends_on:
+      omniroute:
+        condition: service_healthy
+    command: ["node", "/opt/omniroute/docker/devin-bridge/run-contract.mjs"]
+    volumes:
+      - ./run-contract.mjs:/opt/omniroute/docker/devin-bridge/run-contract.mjs:ro
+
+  claude-egress-guard:
+    image: node:26.0.0-bookworm-slim
+    profiles: [offline, live-devin]
+    user: "10001:10001"
+    read_only: true
+    cap_drop: [ALL]
+    security_opt: [no-new-privileges:true]
+    command: ["node", "/guard/proxy.mjs"]
+    environment:
+      GUARD_LISTEN: 0.0.0.0:8080
+      GUARD_POLICY: deny-all
+      GUARD_LOG: /guard-audit/egress.jsonl
+    healthcheck:
+      test:
+        [
+          "CMD",
+          "node",
+          "-e",
+          "require('net').connect(8080,'127.0.0.1').on('connect',()=>process.exit(0)).on('error',()=>process.exit(1))",
+        ]
+      interval: 1s
+      timeout: 1s
+      retries: 15
+    volumes:
+      - ./network-guard:/guard:ro
+      - ../../.sandbox/guard-audit/claude:/guard-audit
+    networks: [bridge-internal]
+
+  network-guard:
+    image: node:26.0.0-bookworm-slim
+    profiles: [live-devin]
+    user: "10001:10001"
+    read_only: true
+    cap_drop: [ALL]
+    security_opt: [no-new-privileges:true]
+    command: ["node", "/guard/proxy.mjs"]
+    environment:
+      GUARD_LISTEN: 0.0.0.0:8080
+      GUARD_POLICY: devin
+      GUARD_LOG: /guard-audit/egress.jsonl
+    healthcheck:
+      test:
+        [
+          "CMD",
+          "node",
+          "-e",
+          "require('net').connect(8080,'127.0.0.1').on('connect',()=>process.exit(0)).on('error',()=>process.exit(1))",
+        ]
+      interval: 1s
+      timeout: 1s
+      retries: 15
+    volumes:
+      - ./network-guard:/guard:ro
+      - ../../.sandbox/guard-audit/devin:/guard-audit
+    networks: [devin-guard-internal, guard-egress]
+
+  omniroute-live:
+    <<: *runtime
+    profiles: [live-devin]
+    hostname: omniroute
+    depends_on:
+      network-guard:
+        condition: service_healthy
+    environment:
+      <<: *isolated-environment
+      CLI_DEVIN_AGENTIC_BIN: /usr/local/bin/devin
+      DEVIN_BRIDGE_PROXY_URL: http://network-guard:8080
+    networks: [bridge-internal, devin-guard-internal]
+    command: ["npm", "run", "start"]
+    healthcheck:
+      test:
+        [
+          "CMD",
+          "node",
+          "-e",
+          "fetch('http://127.0.0.1:20128/healthz').then(r=>process.exit(r.ok?0:1)).catch(()=>process.exit(1))",
+        ]
+      interval: 2s
+      timeout: 2s
+      retries: 60
+    volumes:
+      - devin-auth:/home/bridge/.local/share/devin
+      - omniroute-live-data:/home/bridge/.omniroute-isolated
+
+  claude-live:
+    <<: *runtime
+    profiles: [live-devin]
+    depends_on:
+      omniroute-live:
+        condition: service_healthy
+      claude-egress-guard:
+        condition: service_healthy
+    working_dir: /workspace
+    command: ["bash", "/opt/omniroute/docker/devin-bridge/run-claude-live-e2e.sh"]
+    environment:
+      <<: *isolated-environment
+      NODE_USE_ENV_PROXY: "1"
+      HTTP_PROXY: http://claude-egress-guard:8080
+      HTTPS_PROXY: http://claude-egress-guard:8080
+      NO_PROXY: omniroute
+    volumes:
+      - claude-isolated-config:/home/bridge/.claude-devin-isolated
+      - ../../.sandbox/live-workspace:/workspace
+      - ../../.sandbox/evidence:/evidence
+      - ./run-claude-live-e2e.sh:/opt/omniroute/docker/devin-bridge/run-claude-live-e2e.sh:ro
+
+networks:
+  bridge-internal:
+    internal: true
+  devin-guard-internal:
+    internal: true
+  guard-egress: {}
+
+volumes:
+  claude-isolated-config: {}
+  devin-auth: {}
+  omniroute-offline-data: {}
+  omniroute-live-data: {}

docker/devin-bridge/mock-devin.mjs

@@ -0,0 +1,229 @@
+#!/usr/bin/env node
+import fs from "node:fs";
+import readline from "node:readline";
+
+if (
+  process.argv[2] !== "acp" ||
+  process.argv[3] !== "--agent-type" ||
+  process.argv[4] !== "summarizer" ||
+  process.argv.length !== 5
+) {
+  process.exit(64);
+}
+
+const logFile = process.env.DEVIN_BRIDGE_MOCK_LOG || "/evidence/mock-acp.jsonl";
+const rl = readline.createInterface({ input: process.stdin });
+const send = (value) => process.stdout.write(`${JSON.stringify(value)}\n`);
+const log = (value) => fs.appendFileSync(logFile, `${JSON.stringify(value)}\n`);
+
+const actions = [
+  {
+    name: "Skill",
+    arguments: { skill: "bridge-proof" },
+  },
+  {
+    name: "Bash",
+    arguments: {
+      command: "find . -maxdepth 2 -type f -print",
+      description: "Locate the fixture files",
+    },
+  },
+  {
+    name: "Read",
+    arguments: { file_path: "/workspace/math.js" },
+  },
+  {
+    name: "Edit",
+    arguments: {
+      file_path: "/workspace/math.js",
+      old_string: "return a - b;",
+      new_string: "return a * b;",
+    },
+  },
+  {
+    name: "Bash",
+    arguments: { command: "npm test", description: "Run the fixture tests" },
+  },
+  {
+    name: "Edit",
+    arguments: {
+      file_path: "/workspace/math.js",
+      old_string: "return a * b;",
+      new_string: "return a + b;",
+    },
+  },
+  {
+    name: "Bash",
+    arguments: { command: "npm test", description: "Confirm the corrected fixture" },
+  },
+];
+
+rl.on("line", (line) => {
+  const message = JSON.parse(line);
+  if (message.method === "initialize") {
+    if (message.params?.protocolVersion !== 1) {
+      send({ jsonrpc: "2.0", id: message.id, error: { code: -32602, message: "ACP v1 required" } });
+      return;
+    }
+    send({ jsonrpc: "2.0", id: message.id, result: { protocolVersion: 1 } });
+  } else if (message.method === "session/new") {
+    if (message.params?.cwd !== "/home/bridge" || !Array.isArray(message.params?.mcpServers)) {
+      send({ jsonrpc: "2.0", id: message.id, error: { code: -32602, message: "unsafe session" } });
+      return;
+    }
+    send({
+      jsonrpc: "2.0",
+      id: message.id,
+      result: { sessionId: "offline" },
+    });
+  } else if (message.method === "session/set_config_option") {
+    send({
+      jsonrpc: "2.0",
+      id: message.id,
+      error: { code: -32602, message: "summarizer mode must not be mutated" },
+    });
+  } else if (message.method === "session/prompt") {
+    const prompt = String(message.params?.prompt?.[0]?.text || "");
+    if (!prompt.includes("[Devin Summarizer Bridge]") || !prompt.includes("[Execution Trace]")) {
+      send({
+        jsonrpc: "2.0",
+        id: message.id,
+        error: { code: -32602, message: "summarizer bridge framing required" },
+      });
+      return;
+    }
+    if (prompt.includes("CONTRACT_AFTER_TOOL")) {
+      log({ provider: "devin-cli-agentic", scenario: "after-tool" });
+      send({
+        jsonrpc: "2.0",
+        method: "session/update",
+        params: {
+          sessionId: "offline",
+          update: {
+            sessionUpdate: "agent_message_chunk",
+            content: { type: "text", text: "contract continued" },
+          },
+        },
+      });
+      send({ jsonrpc: "2.0", id: message.id, result: { stopReason: "end_turn" } });
+      return;
+    }
+    if (prompt.includes("CONTRACT_EXIT")) {
+      log({ provider: "devin-cli-agentic", scenario: "exit" });
+      process.exit(7);
+    }
+    if (prompt.includes("CONTRACT_ERROR")) {
+      log({ provider: "devin-cli-agentic", scenario: "error" });
+      send({
+        jsonrpc: "2.0",
+        id: message.id,
+        error: { code: -32000, message: "deterministic upstream failure" },
+      });
+      return;
+    }
+    if (prompt.includes("CONTRACT_TEXT")) {
+      log({ provider: "devin-cli-agentic", scenario: "text" });
+      send({
+        jsonrpc: "2.0",
+        method: "session/update",
+        params: {
+          sessionId: "offline",
+          update: {
+            sessionUpdate: "agent_message_chunk",
+            content: { type: "text", text: "contract text" },
+          },
+        },
+      });
+      send({ jsonrpc: "2.0", id: message.id, result: { stopReason: "end_turn" } });
+      return;
+    }
+    if (prompt.includes("CONTRACT_NARRATIVE_REPAIR")) {
+      const isRepair = prompt.includes("[Single Repair Attempt]");
+      log({
+        provider: "devin-cli-agentic",
+        scenario: "narrative-repair",
+        stage: isRepair ? "repair" : "initial",
+      });
+      send({
+        jsonrpc: "2.0",
+        method: "session/update",
+        params: {
+          sessionId: "offline",
+          update: {
+            sessionUpdate: "agent_message_chunk",
+            content: {
+              type: "text",
+              text: isRepair
+                ? '<tool>{"name":"Read","arguments":{"file_path":"/workspace/math.js"}}</tool>'
+                : "I'll start by reading the math.js file, then run the tests.",
+            },
+          },
+        },
+      });
+      send({ jsonrpc: "2.0", id: message.id, result: { stopReason: "end_turn" } });
+      return;
+    }
+    if (prompt.includes("CONTRACT_TOOL")) {
+      log({ provider: "devin-cli-agentic", scenario: "tool" });
+      send({
+        jsonrpc: "2.0",
+        method: "session/update",
+        params: {
+          sessionId: "offline",
+          update: {
+            sessionUpdate: "agent_message_chunk",
+            content: {
+              type: "text",
+              text: '<tool>{"name":"Read","arguments":{"file_path":"/workspace/math.js"}}</tool>',
+            },
+          },
+        },
+      });
+      send({ jsonrpc: "2.0", id: message.id, result: { stopReason: "end_turn" } });
+      return;
+    }
+    const resultCount = (prompt.match(/\[Tool Result\]/g) || []).length;
+    if (!prompt.includes("CLAUDE_MD_BRIDGE_ACTIVE") || !prompt.includes("COMMAND_BRIDGE_ACTIVE")) {
+      send({
+        jsonrpc: "2.0",
+        id: message.id,
+        error: { code: -32602, message: "Claude project context missing" },
+      });
+      return;
+    }
+
+    const action = actions[resultCount];
+    const text = action
+      ? `<tool>${JSON.stringify(action)}</tool>`
+      : "BRIDGE_E2E_COMPLETE CLAUDE_MD_BRIDGE_ACTIVE SKILL_BRIDGE_ACTIVE COMMAND_BRIDGE_ACTIVE";
+    if (!action && !prompt.includes("SKILL_BRIDGE_ACTIVE")) {
+      send({
+        jsonrpc: "2.0",
+        id: message.id,
+        error: { code: -32602, message: "Skill result missing" },
+      });
+      return;
+    }
+    log({
+      provider: "devin-cli-agentic",
+      model: message.params?.model || "swe-1-7",
+      resultCount,
+      action: action?.name || "final",
+    });
+    const midpoint = Math.max(1, Math.floor(text.length / 2));
+    for (const chunk of [text.slice(0, midpoint), text.slice(midpoint)]) {
+      send({
+        jsonrpc: "2.0",
+        method: "session/update",
+        params: {
+          sessionId: "offline",
+          update: {
+            sessionUpdate: "agent_message_chunk",
+            content: { type: "text", text: chunk },
+          },
+        },
+      });
+    }
+    send({ jsonrpc: "2.0", id: message.id, result: { stopReason: "end_turn" } });
+  }
+});

docker/devin-bridge/network-guard/policy.mjs

@@ -0,0 +1,130 @@
+export const DEVIN_ALLOWED_SUFFIXES = Object.freeze([".devin.ai", ".cognition.ai"]);
+export const DEVIN_ALLOWED_EXACT_HOSTS = Object.freeze([
+  "server.codeium.com",
+  "unleash.codeium.com",
+]);
+
+function normalizeHostname(hostname) {
+  return String(hostname || "")
+    .trim()
+    .toLowerCase()
+    .replace(/\.$/, "");
+}
+
+export function isAllowedGuardHostname(hostname, policy = "deny-all") {
+  if (policy !== "devin") return false;
+  const value = normalizeHostname(hostname);
+  if (!value) return false;
+  if (DEVIN_ALLOWED_EXACT_HOSTS.includes(value)) return true;
+  return DEVIN_ALLOWED_SUFFIXES.some(
+    (suffix) => value === suffix.slice(1) || value.endsWith(suffix)
+  );
+}
+
+const HOP_BY_HOP_HEADERS = new Set([
+  "connection",
+  "keep-alive",
+  "proxy-authenticate",
+  "proxy-authorization",
+  "proxy-connection",
+  "te",
+  "trailer",
+  "transfer-encoding",
+  "upgrade",
+]);
+
+export function sanitizeForwardHeaders(headers, target) {
+  const connectionTokens = String(headers.connection || "")
+    .split(",")
+    .map((value) => value.trim().toLowerCase())
+    .filter(Boolean);
+  const blocked = new Set([...HOP_BY_HOP_HEADERS, ...connectionTokens]);
+  const sanitized = {};
+  for (const [name, value] of Object.entries(headers)) {
+    if (value === undefined || blocked.has(name.toLowerCase()) || name.toLowerCase() === "host") {
+      continue;
+    }
+    sanitized[name] = value;
+  }
+  sanitized.host = target.host;
+  return sanitized;
+}
+
+export function parseConnectAuthority(authority) {
+  const value = String(authority || "");
+  const match = value.match(/^(?:\[([^\]]+)\]|([^:]+)):(\d+)$/);
+  if (!match) return null;
+  const hostname = normalizeHostname(match[1] || match[2]);
+  const port = Number(match[3]);
+  if (!hostname || port !== 443) return null;
+  return { hostname, port };
+}
+
+function readUint24(buffer, offset) {
+  return (buffer[offset] << 16) | (buffer[offset + 1] << 8) | buffer[offset + 2];
+}
+
+export function parseTlsClientHelloSni(buffer) {
+  if (!Buffer.isBuffer(buffer)) return { status: "invalid", reason: "not_buffer" };
+  let offset = 0;
+  const handshakeParts = [];
+  while (offset < buffer.length) {
+    if (buffer.length - offset < 5) return { status: "need-more" };
+    if (buffer[offset] !== 22) return { status: "invalid", reason: "not_handshake_record" };
+    const recordLength = buffer.readUInt16BE(offset + 3);
+    if (recordLength <= 0 || recordLength > 18432) {
+      return { status: "invalid", reason: "invalid_record_length" };
+    }
+    if (buffer.length - offset - 5 < recordLength) return { status: "need-more" };
+    handshakeParts.push(buffer.subarray(offset + 5, offset + 5 + recordLength));
+    offset += 5 + recordLength;
+  }
+  const handshake = Buffer.concat(handshakeParts);
+  if (handshake.length < 4) return { status: "need-more" };
+  if (handshake[0] !== 1) return { status: "invalid", reason: "not_client_hello" };
+  const helloLength = readUint24(handshake, 1);
+  if (helloLength > 65531) return { status: "invalid", reason: "client_hello_too_large" };
+  if (handshake.length - 4 < helloLength) return { status: "need-more" };
+  const hello = handshake.subarray(4, 4 + helloLength);
+  let cursor = 34;
+  if (hello.length < cursor + 1) return { status: "invalid", reason: "truncated_hello" };
+  const sessionLength = hello[cursor++];
+  cursor += sessionLength;
+  if (hello.length < cursor + 2) return { status: "invalid", reason: "truncated_ciphers" };
+  const cipherLength = hello.readUInt16BE(cursor);
+  cursor += 2 + cipherLength;
+  if (hello.length < cursor + 1) return { status: "invalid", reason: "truncated_compression" };
+  const compressionLength = hello[cursor++];
+  cursor += compressionLength;
+  if (hello.length < cursor + 2) return { status: "invalid", reason: "missing_extensions" };
+  const extensionsLength = hello.readUInt16BE(cursor);
+  cursor += 2;
+  const extensionsEnd = cursor + extensionsLength;
+  if (extensionsEnd > hello.length) return { status: "invalid", reason: "truncated_extensions" };
+  while (cursor < extensionsEnd) {
+    if (extensionsEnd - cursor < 4) return { status: "invalid", reason: "truncated_extension" };
+    const type = hello.readUInt16BE(cursor);
+    const length = hello.readUInt16BE(cursor + 2);
+    cursor += 4;
+    if (cursor + length > extensionsEnd) {
+      return { status: "invalid", reason: "invalid_extension_length" };
+    }
+    if (type === 0) {
+      const data = hello.subarray(cursor, cursor + length);
+      if (data.length < 5 || data.readUInt16BE(0) !== data.length - 2 || data[2] !== 0) {
+        return { status: "invalid", reason: "invalid_server_name" };
+      }
+      const nameLength = data.readUInt16BE(3);
+      if (nameLength !== data.length - 5) {
+        return { status: "invalid", reason: "invalid_server_name_length" };
+      }
+      const serverName = normalizeHostname(data.subarray(5).toString("ascii"));
+      if (!/^[a-z0-9.-]+$/.test(serverName)) {
+        return { status: "invalid", reason: "invalid_server_name_value" };
+      }
+      return { status: "ok", serverName };
+    }
+    cursor += length;
+  }
+  return { status: "invalid", reason: "missing_sni" };
+}

docker/devin-bridge/network-guard/proxy.mjs

@@ -0,0 +1,136 @@
+import fs from "node:fs";
+import http from "node:http";
+import net from "node:net";
+import { pathToFileURL } from "node:url";
+
+import {
+  isAllowedGuardHostname,
+  parseConnectAuthority,
+  parseTlsClientHelloSni,
+  sanitizeForwardHeaders,
+} from "./policy.mjs";
+
+const MAX_CLIENT_HELLO_BYTES = 64 * 1024;
+const CLIENT_HELLO_TIMEOUT_MS = 3000;
+
+export function createGuardProxy({
+  policy = "deny-all",
+  logPath = "/tmp/egress.jsonl",
+  allowHostname = (hostname) => isAllowedGuardHostname(hostname, policy),
+  connectSocket = (port, hostname, onConnect) => net.connect(port, hostname, onConnect),
+} = {}) {
+  if (!new Set(["deny-all", "devin"]).has(policy)) {
+    throw new Error(`Unknown network guard policy: ${policy}`);
+  }
+
+  function audit(hostname, decision, reason) {
+    fs.appendFileSync(
+      logPath,
+      `${JSON.stringify({ at: new Date().toISOString(), hostname, decision, reason })}\n`
+    );
+  }
+
+  const server = http.createServer((req, res) => {
+    let target;
+    try {
+      target = new URL(req.url);
+    } catch {
+      res.writeHead(400).end("invalid proxy target\n");
+      return;
+    }
+    if (target.protocol !== "http:" || target.username || target.password) {
+      audit(target.hostname, "deny", "invalid_http_target");
+      res.writeHead(403).end("egress denied\n");
+      return;
+    }
+    if (!allowHostname(target.hostname)) {
+      audit(target.hostname, "deny", "host_policy");
+      res.writeHead(403).end("egress denied\n");
+      return;
+    }
+    audit(target.hostname, "allow", "host_policy");
+    const upstream = http.request(
+      target,
+      {
+        method: req.method,
+        headers: sanitizeForwardHeaders(req.headers, target),
+      },
+      (reply) => {
+        res.writeHead(reply.statusCode || 502, reply.headers);
+        reply.pipe(res);
+      }
+    );
+    req.pipe(upstream);
+    upstream.on("error", () => res.writeHead(502).end("upstream error\n"));
+  });
+
+  server.on("connect", (req, client, head) => {
+    const authority = parseConnectAuthority(req.url);
+    if (!authority) {
+      audit(req.url, "deny", "invalid_connect_authority");
+      client.end("HTTP/1.1 403 Forbidden\r\n\r\n");
+      return;
+    }
+    const { hostname, port } = authority;
+    if (!allowHostname(hostname)) {
+      audit(hostname, "deny", "host_policy");
+      client.end("HTTP/1.1 403 Forbidden\r\n\r\n");
+      return;
+    }
+
+    let buffer = Buffer.from(head);
+    let settled = false;
+    const timer = setTimeout(() => fail("client_hello_timeout"), CLIENT_HELLO_TIMEOUT_MS);
+    timer.unref?.();
+
+    const cleanup = () => {
+      clearTimeout(timer);
+      client.removeListener("data", onData);
+    };
+    const fail = (reason) => {
+      if (settled) return;
+      settled = true;
+      cleanup();
+      audit(hostname, "deny", reason);
+      client.destroy();
+    };
+    const inspect = () => {
+      if (buffer.length > MAX_CLIENT_HELLO_BYTES) return fail("client_hello_too_large");
+      const parsed = parseTlsClientHelloSni(buffer);
+      if (parsed.status === "need-more") return;
+      if (parsed.status !== "ok") return fail(parsed.reason || "invalid_client_hello");
+      if (parsed.serverName !== hostname) return fail("sni_mismatch");
+      settled = true;
+      cleanup();
+      client.pause();
+      const upstream = connectSocket(port, hostname, () => {
+        audit(hostname, "allow", "sni_match");
+        if (buffer.length) upstream.write(buffer);
+        upstream.pipe(client);
+        client.pipe(upstream);
+        client.resume();
+      });
+      upstream.on("error", () => client.destroy());
+    };
+    const onData = (chunk) => {
+      buffer = Buffer.concat([buffer, chunk]);
+      inspect();
+    };
+
+    client.write("HTTP/1.1 200 Connection Established\r\n\r\n");
+    client.on("data", onData);
+    if (buffer.length) inspect();
+    client.resume();
+  });
+
+  return server;
+}
+
+if (process.argv[1] && pathToFileURL(process.argv[1]).href === import.meta.url) {
+  const [host, portText] = (process.env.GUARD_LISTEN || "0.0.0.0:8080").split(":");
+  const server = createGuardProxy({
+    policy: process.env.GUARD_POLICY || "deny-all",
+    logPath: process.env.GUARD_LOG || "/tmp/egress.jsonl",
+  });
+  server.listen(Number(portText), host);
+}

docker/devin-bridge/run-claude-e2e.sh

@@ -0,0 +1,27 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+unset ANTHROPIC_API_KEY CLAUDE_CODE_OAUTH_TOKEN ANTHROPIC_BEDROCK_BASE_URL ANTHROPIC_VERTEX_BASE_URL
+unset CLAUDE_CODE_USE_BEDROCK CLAUDE_CODE_USE_VERTEX CLAUDE_CODE_USE_FOUNDRY
+
+set -o pipefail
+check() {
+  "$@"
+  printf 'E2E check passed: %s\n' "$*"
+}
+
+claude -p --output-format stream-json --verbose --max-turns 12 \
+  --permission-mode bypassPermissions \
+  "/bridge-check" | tee /evidence/claude-stream.jsonl
+
+if grep -Eqi 'log[ -]?in|authenticate.*anthropic|claude\.ai' /evidence/claude-stream.jsonl; then
+  echo "Claude Code requested forbidden authentication" >&2
+  exit 1
+fi
+check grep -q 'return a + b;' /workspace/math.js
+npm test
+check grep -q 'Skill' /workspace/.e2e-hook.log
+check grep -q 'Read' /workspace/.e2e-hook.log
+check grep -q 'Edit' /workspace/.e2e-hook.log
+check grep -q 'Bash' /workspace/.e2e-hook.log
+check grep -q 'BRIDGE_E2E_COMPLETE' /evidence/claude-stream.jsonl

docker/devin-bridge/run-claude-live-e2e.sh

@@ -0,0 +1,52 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+unset ANTHROPIC_API_KEY CLAUDE_CODE_OAUTH_TOKEN ANTHROPIC_BEDROCK_BASE_URL ANTHROPIC_VERTEX_BASE_URL
+unset CLAUDE_CODE_USE_BEDROCK CLAUDE_CODE_USE_VERTEX CLAUDE_CODE_USE_FOUNDRY
+
+bridge_system_prompt="You are a coding agent inside Claude Code. Use only the client-owned tools supplied in the request. Never execute or request a Devin-owned tool. When work requires a tool, select the appropriate client tool and wait for its result before continuing."
+scenario_cooldown_seconds="${DEVIN_BRIDGE_LIVE_SCENARIO_COOLDOWN_SECONDS:-15}"
+
+run_scenario() {
+  local evidence_file="$1"
+  local prompt="$2"
+  claude -p --output-format stream-json --verbose --max-turns 12 \
+    --tools Read,Edit,Bash \
+    --system-prompt "$bridge_system_prompt" \
+    --permission-mode bypassPermissions "$prompt" | tee "$evidence_file"
+  if grep -Eqi 'log[ -]?in|authenticate.*anthropic|claude\.ai' "$evidence_file"; then
+    echo "Claude Code requested forbidden authentication" >&2
+    exit 1
+  fi
+}
+
+validate_scenario() {
+  local evidence_file="$1"
+  local marker="$2"
+  local required_tools="$3"
+  local require_npm_test="$4"
+  local required_slash_command="${5:-}"
+  local required_skill="${6:-}"
+  local accept_explicit_completion="${7:-false}"
+  node /opt/omniroute/scripts/devin-bridge/validate-claude-evidence.mjs \
+    "$evidence_file" "$marker" "$required_tools" "$require_npm_test" \
+    "$required_slash_command" "$required_skill" "$accept_explicit_completion"
+}
+
+run_scenario /evidence/live-analysis.jsonl \
+  "Read /workspace/CLAUDE.md, /workspace/math.js, and /workspace/math.test.js directly without searching or editing. Explain the defect, then end with LIVE_ANALYSIS_COMPLETE."
+validate_scenario /evidence/live-analysis.jsonl LIVE_ANALYSIS_COMPLETE Read false
+sleep "$scenario_cooldown_seconds"
+
+run_scenario /evidence/live-fix.jsonl \
+  "Use Edit now to replace 'return a - b;' with 'return a + b;' in /workspace/math.js. Then use Bash to run npm test. Do not summarize before npm test succeeds. End with LIVE_FIX_COMPLETE only after the test passes."
+grep -q 'return a + b;' /workspace/math.js
+npm test
+validate_scenario /evidence/live-fix.jsonl LIVE_FIX_COMPLETE Edit,Bash true
+sleep "$scenario_cooldown_seconds"
+
+run_scenario /evidence/live-command.jsonl "/bridge-check"
+validate_scenario /evidence/live-command.jsonl BRIDGE_E2E_COMPLETE Bash true \
+  bridge-check bridge-proof true
+
+printf 'PASS: three live Devin-backed Claude Code scenarios completed\n'

docker/devin-bridge/run-contract.mjs

@@ -0,0 +1,135 @@
+#!/usr/bin/env node
+import assert from "node:assert/strict";
+
+const endpoint = "http://omniroute:20128/v1/messages";
+const headers = {
+  "anthropic-version": "2023-06-01",
+  "content-type": "application/json",
+  "x-api-key": "sk-local-devin-gateway",
+};
+const model = process.env.DEVIN_BRIDGE_MODEL || "devin-cli-agentic/swe-1-7";
+
+async function request(prompt, extra = {}) {
+  return fetch(endpoint, {
+    method: "POST",
+    headers,
+    body: JSON.stringify({
+      model,
+      max_tokens: 256,
+      messages: [{ role: "user", content: prompt }],
+      ...extra,
+    }),
+  });
+}
+
+const textReply = await request("CONTRACT_TEXT");
+assert.equal(textReply.status, 200);
+assert.match(textReply.headers.get("content-type") || "", /application\/json/);
+const textBody = await textReply.json();
+assert.equal(textBody.type, "message");
+assert.equal(textBody.role, "assistant");
+assert.equal(textBody.stop_reason, "end_turn");
+assert.deepEqual(textBody.content, [{ type: "text", text: "contract text" }]);
+
+const toolReply = await request("CONTRACT_TOOL", {
+  stream: true,
+  tools: [
+    {
+      name: "Read",
+      description: "Read a file",
+      input_schema: {
+        type: "object",
+        properties: { file_path: { type: "string" } },
+        required: ["file_path"],
+        additionalProperties: false,
+      },
+    },
+  ],
+});
+assert.equal(toolReply.status, 200);
+assert.match(toolReply.headers.get("content-type") || "", /text\/event-stream/);
+const toolStream = await toolReply.text();
+const eventNames = toolStream
+  .split("\n")
+  .filter((line) => line.startsWith("event: "))
+  .map((line) => line.slice(7));
+assert.deepEqual(eventNames, [
+  "message_start",
+  "content_block_start",
+  "content_block_delta",
+  "content_block_stop",
+  "message_delta",
+  "message_stop",
+]);
+const toolEvents = toolStream
+  .split("\n")
+  .filter((line) => line.startsWith("data: "))
+  .map((line) => JSON.parse(line.slice(6)));
+const toolUse = toolEvents.find((event) => event.type === "content_block_start")?.content_block;
+assert.equal(toolUse?.type, "tool_use");
+assert.equal(toolUse?.name, "Read");
+assert.match(toolUse?.id || "", /^tool_devin_/);
+
+const repairedNarrativeReply = await request("CONTRACT_NARRATIVE_REPAIR", {
+  tools: [
+    {
+      name: "Read",
+      description: "Read a file",
+      input_schema: {
+        type: "object",
+        properties: { file_path: { type: "string" } },
+        required: ["file_path"],
+        additionalProperties: false,
+      },
+    },
+  ],
+});
+assert.equal(repairedNarrativeReply.status, 200);
+const repairedNarrativeBody = await repairedNarrativeReply.json();
+assert.equal(repairedNarrativeBody.stop_reason, "tool_use");
+assert.equal(repairedNarrativeBody.content?.[0]?.type, "tool_use");
+assert.equal(repairedNarrativeBody.content?.[0]?.name, "Read");
+
+const continuationReply = await fetch(endpoint, {
+  method: "POST",
+  headers,
+  body: JSON.stringify({
+    model,
+    max_tokens: 256,
+    tools: [
+      {
+        name: "Read",
+        description: "Read a file",
+        input_schema: { type: "object", properties: {}, additionalProperties: true },
+      },
+    ],
+    messages: [
+      { role: "user", content: "CONTRACT_TOOL" },
+      { role: "assistant", content: [toolUse] },
+      {
+        role: "user",
+        content: [
+          {
+            type: "tool_result",
+            tool_use_id: toolUse.id,
+            content: "CONTRACT_AFTER_TOOL",
+          },
+        ],
+      },
+    ],
+  }),
+});
+assert.equal(continuationReply.status, 200);
+const continuationBody = await continuationReply.json();
+assert.equal(continuationBody.stop_reason, "end_turn");
+assert.deepEqual(continuationBody.content, [{ type: "text", text: "contract continued" }]);
+
+for (const marker of ["CONTRACT_ERROR", "CONTRACT_EXIT"]) {
+  const failedReply = await request(marker);
+  assert.equal(failedReply.status, 502);
+  const failedBody = await failedReply.json();
+  assert.equal(failedBody.error?.type, "server_error");
+  assert.doesNotMatch(JSON.stringify(failedBody), /stack|anthropic|openai/i);
+}
+
+console.log("PASS: Anthropic Messages wire contracts and fail-closed errors passed");