next@16 lists @swc/helpers@0.5.15 in its own dependencies but npm's
deduplication during global install fails to place it in the omniroute
app's node_modules when hoisted. This causes MODULE_NOT_FOUND for
@swc/helpers/esm/_interop_require_default.js on startup.
Fix: add @swc/helpers@0.5.19 to omniroute's top-level dependencies and
overrides so npm guarantees its presence regardless of hoisting strategy.
Reproducible on Windows (Node 22) and Linux.
When all provider quotas are exhausted (reservoir=0 after repeated 429s),
Bottleneck's schedule() would queue requests indefinitely since no maxWait
was configured. Clients (Cursor, Claude Code, VS Code) would hang forever.
Fix: add maxWait=120000 (2min, configurable via RATE_LIMIT_MAX_WAIT_MS env)
to DEFAULT_SETTINGS and all three Bottleneck constructors. When a job waits
longer than maxWait, Bottleneck rejects with a BottleneckError which
propagates as a 502/503 error to the client — a clean fail-fast instead
of infinite hang.
The healthcheck script was querying /api/settings which returns config
data rather than system health. Updated to /api/monitoring/health which
is the canonical health endpoint used across tests, SystemMonitor.tsx,
MaintenanceBanner.tsx, playwright config, and MCP tools.
OpenAI-compatible providers (OpenAI, Codex) reject name:'' with 400 errors:
- 'Unknown parameter: input[1].name'
- 'Invalid tools[0].name: empty string'
Some clients (e.g. PocketPaw) forward assistant turns with name:'' in
the OpenAI Responses API input[] and chat completions messages[].
Fix: filter out name:'' from messages[] and input[] before translateRequest.
Non-empty non-null name values are preserved per OpenAI spec.
When proxying Claude responses through OmniRoute, thinking blocks were being
emitted as regular content (delta.content) with <think>...</think> XML tags.
Clients like Claude Code, Cursor, and Windsurf look for delta.reasoning_content
to render the thinking panel — not <think> tags inside content.
Root cause (claude-to-openai.ts):
- content_block_start type:thinking → emitted { content: '<think>' }
- content_block_delta thinking_delta → emitted { content: delta.thinking }
- content_block_stop thinking block → emitted { content: '</think>' }
Fix:
- content_block_start → emits { reasoning_content: '' } (signals block start)
- thinking_delta → emits { reasoning_content: delta.thinking }
- content_block_stop → no extra chunk needed (thinking streamed via reasoning_content)
This fix applies when sourceFormat=CLAUDE targetFormat=OPENAI (Antigravity OAuth,
direct Claude API providers). The user reported 'Thinking Budget: passthrough'
was enabled but thinking was invisible — this is the root cause.
Fixes#289
Resolves root cause of #252 (Electron black screen) and #249 (OAuth fail)
for users running with zero configuration (no .env needed).
New: scripts/bootstrap-env.mjs
- Auto-generates JWT_SECRET (64 bytes), STORAGE_ENCRYPTION_KEY (32 bytes),
API_KEY_SECRET (32 bytes) if missing or empty
- Persists to {DATA_DIR}/server.env — survives restarts, Docker volume
remounts, and upgrades without changing secrets
- Reads .env from CWD (user overrides), then merges process.env (highest prio)
- Logs friendly warnings for missing optional OAuth secrets
Updated: run-standalone.mjs + run-next.mjs
- Call bootstrapEnv() before spawning server — covers npm + Docker paths
Updated: electron/main.js (synchronous inline — CJS cannot await import ESM)
- Reads userData/server.env, generates missing secrets with crypto.randomBytes()
- Persists back to server.env, sets OMNIROUTE_BOOTSTRAPPED=true
New: BootstrapBanner.tsx + page.tsx update
- Dismissable amber banner on dashboard home when running in zero-config mode
- Shows where server.env is located and how to customize secrets
In packaged Electron on macOS/Windows/Linux, there is no .env file.
The Next.js server needs JWT_SECRET and STORAGE_ENCRYPTION_KEY to start —
without them it crashes silently, causing ERR_CONNECTION_REFUSED
and a black screen in the Electron window.
Fix: Generate cryptographically random values with crypto.randomBytes()
on first launch, persist them in userData/electron-env.json, and pass
them to the spawned server.js process via the env option.
Root cause: macOS users reported 'app black screen' (#252) and
ERR_CONNECTION_REFUSED — this was the Next.js server crashing at startup
because these env vars don't exist in the desktop OS environment.
- Step 4 now marked ⚠️ MANDATORY with CI will fail warning
- Command is now auto-extracting version from package.json (no manual substitution)
- Step 4 has // turbo annotation for auto-execution
- Added 'Known CI Pitfalls' table: docs-sync failures, Electron fpm, Docker 502
check:docs-sync fails when openapi.yaml version != package.json version.
Updating to match after v2.2.4 release.
Systematic fix: openapi.yaml version must always be updated alongside
package.json during releases (see generate-release workflow step 4).
CI Lint fixes:
- docs/openapi.yaml: bump version 2.2.0 → 2.2.3 (was out of sync with package.json)
- CHANGELOG.md: add '## [Unreleased]' as first section (required by check:docs-sync)
Electron Linux fix:
- electron-release.yml: add 'gem install fpm' step for Linux builds
fpm is required by electron-builder to package .deb installers;
ubuntu-latest runners don't have it pre-installed
Docker publish:
- docker-publish.yml: add DOCKER_BUILDKIT_INLINE_CACHE env; prev 502 was
a transient Docker Hub network error, no code change needed
Users are now directed to run 'npm run system-info' when reporting bugs.
Added to:
- ## Support → '🐛 Reporting a Bug?' subsection
- Pain point #10 '🐛 I can't diagnose errors' bullet list
- New npm run system-info command (#280)
- Kimi K2/K2.5 Fireworks path aliases (#265)
- Mistral/Llama short aliases (#278)
- Custom alias instructions via Settings → Model Aliases
npm ci fails if the tag commit's lock file is out of sync (as happened
with v2.2.0 when @swc/helpers was missing). npm install is safe here
because the publish workflow only needs deps to run prepublish.mjs —
strict lock enforcement is not required for the publish step.
Prevents E403 failures when a release event fires more than once for the
same version (e.g. re-running a failed workflow or duplicate tag event).
The publish step now checks whether the version is already on npm and
exits cleanly with a warning instead of failing the workflow.
All 3 CI workflows (CI, npm-publish, docker-publish) were failing with:
'Missing: @swc/helpers@0.5.19 from lock file'
Root cause: npm version bump ran without npm install, causing the lock file
to be out of sync with CI's npm resolver (which resolves @swc/helpers@0.5.19
while local npm 10.9.4 resolves 0.5.15).
Fix: added 'overrides': { '@swc/helpers': '^0.5.19' } to package.json and
regenerated package-lock.json with npm install.
Also updated .agents/workflows/generate-release.md to enforce:
- Always use 'npm version patch --no-git-tag-version' (never minor/major)
- Always run 'npm install' after bumping to keep lock file in sync
- Version threshold: 2.x.10 → 2.(x+1).0 (manual)
- Merge PR #275: fix(cursor): stabilize tool result handoff
- Merge PR #281: fix(validation): ElevenLabs + Inworld API key validation
- Merge PR #279: feat(kimi-coding): Kimi Coding plan quota display
- Merge PR #264: deps: bump dev group with 5 updates
- Merge PR #263: deps: bump prod group with 2 updates
- fix(ci): repair unit, e2e and i18n test failures
- fix(providers): Add Connection button for OpenAI-compatible providers (#272)
- docs: update CHANGELOG and bump openapi.yaml to 2.2.0
When a manually created OpenAI/Anthropic-compatible provider has 0 connections,
the Add Connection button was hidden behind !isCompatible guard, making it
impossible to add a key. Button now appears in the Connections card header
for compatible providers too, limited to 1 connection (single key per node).
- tests/unit/combo-circuit-breaker: use full model path keys
(combo:groq/llama-3.3-70b) matching combo.ts implementation
- tests/e2e/protocol-visibility: click Protocols tab before asserting
MCP/A2A links (they moved into tab panel, not header nav)
- src/i18n/messages/en.json: add missing header.mcp / header.a2a keys
that caused MISSING_MESSAGE runtime errors during E2E boot
- Replace O(n*k) chunks.some(includes) with emittedToolCallIds Set
- Replace O(n) toolCalls.find with finalizedIds Set (JSON + SSE paths)
- Add byte guard (0x7b check) before payload.toString on every frame
- Add escapeXml() to prevent tag injection in tool result blocks
- Reuse module-level TextDecoder in cursorProtobuf instead of per-call
- Replace [...Set].every() spread with for..of loop
