Files
OmniRoute/docs/security/STEALTH_GUIDE.md
Diego Rodrigues de Sa e Souza 7d57d9f4a1 fix(providers): retire common ChatGPT Web provider (#11754)
Rebased onto the current release/v3.8.51 tip as part of a combined provider-retirement/provenance merge batch (Designer Web, Felo Web, Runtime, GPL-derived removal, Qwen Web already landed). Large conflict set (this is the biggest PR in the batch — the common ChatGPT Web provider touches chat, images, count-tokens, session leases, and combos). Conflicts resolved:

- `open-sse/config/providers/registry/chatgpt-web/*`, `open-sse/executors/chatgpt-web*`, `open-sse/handlers/imageGeneration/providers/chatgptWeb.ts`, and their tests: kept deleted, matching the PR's stated scope.
- `open-sse/config/providers/registry/minimax/web/index.ts`, `open-sse/handlers/imageGeneration/providers/geminiWeb.ts`, `open-sse/executors/gemini-web.ts`'s stale image-mode branch: base-drift collisions against already-merged sibling retirements (#11691, #11708) — kept deleted / dropped the dead code, since this PR's own branch forked before those merged.
- `src/shared/constants/reservedProviderPrefixes.ts`, `open-sse/executors/index.ts`, `executorProxy.ts`, `virtualFactory.ts`, `autoStrategy.ts`, `src/lib/db/providers.ts`, `src/sse/handlers/chat.ts`: combined the Designer + Runtime (Felo/Qwen) + common-ChatGPT-Web retirement guard calls at each shared chokepoint — compute-once-then-OR pattern, consistent with prior combinations in this batch.
- `src/sse/services/model.ts` / `src/sse/handlers/chatHelpers.ts`: adopted this PR's new `getModelInfoOrRetirementResponse()` central wrapper (a real improvement over ad-hoc try/catch), and extended it to also catch the Designer + Runtime retirement errors it didn't originally cover, so the consolidation doesn't regress the other two mechanisms.
- `src/app/api/v1/images/edits/route.ts`: this PR moved the retirement check earlier (before `enforceApiKeyPolicy`) but left the old later call+catch block in place from base drift — removed the now-redundant duplicate `resolveImageRouteModel()` call and merged the Designer catch into the earlier one.
- `open-sse/config/imageRegistry.ts`, `tests/snapshots/executors/executor-map.json` (`keyCount` recomputed to 133), `tests/snapshots/provider/translate-path.json`: same "both sides inserted a different retired provider at the same slot" pattern — resolved by dropping both.
- `tests/unit/chatcore-executor-proxy.test.ts`, `provider-node-reserved-prefix.test.ts`, `combo-auto-candidate-expansion.test.ts`, `messages-count-tokens-route.test.ts`, `virtual-auto-combo.test.ts`: split into independent per-mechanism test blocks (established pattern); `virtual-auto-combo.test.ts`'s old "includes cookie web-session providers" positive-inclusion test (which used chatgpt-web as its example) was retired along with the provider and replaced by this PR's negative-exclusion test for the same slot.
- `docs/architecture/ARCHITECTURE.md`, `CODEBASE_DOCUMENTATION.md` (+ 4 i18n mirrors), `README.md`, `FREE-TIERS-GUIDE.md`, `docs/diagrams/free-tier-budget.svg`, `docs/screenshots/free-tier-budget-card.svg`, `docs/reference/PROVIDER_REFERENCE.md`: recomputed every stale count from the real merged state — 104 executors (`countFiles` gate logic), 351 providers (regenerated via `gen:provider-reference`), 152/351 `hasFree` entries, 445/438/7 free-tier catalog rows, 13 ToS-avoid providers, budget-card regenerated via its real generator script. One doc conflict (`oauth/` module list) needed picking HEAD's side specifically — theirs still listed the already-removed `raycast` module instead of the real `openference`.
- `config/quality/test-masking-allowlist.json`: additive merge of the PR's 17 `_deletedWithReplacement` entries alongside the batch's existing ones (one real duplicate-key mistake in my first pass, caught and fixed via a `object_pairs_hook` duplicate-key check before finalizing).

Also fixed two real, unrelated-to-my-merge issues surfaced by the focused suite:
- `tests/unit/resolve-web-provider-host.test.ts`: the PR's own test had a typo — it asserted `perplexity-web`'s resolved host as `"perplexity.ai"`, but the provider's registered `website` is `"https://www.perplexity.ai"` and the resolver returns the URL's `host` verbatim (no www-stripping), so the correct value is `"www.perplexity.ai"` (consistent with the same test's own `url` assertion).
- `tests/unit/hard-session-lease-bypass-inventory.test.ts`: this golden call-site inventory was already stale on the pristine post-#11713 tip (confirmed via a throwaway probe worktree) — `src/lib/db/providers.ts`'s 3 connection-fallback sites and a third `src/app/api/providers/route.ts` site were never added to the golden list by the earlier-merged #11698/#11720 PRs. Updated it to the real current inventory (dated inline comments explain each delta and which PR introduced it), plus this PR's own legitimate deltas (image-edits duplicate-call removal, `ChatGptWebExecutor.execute()` site removed).

Focused suite green (433/433 across executor-proxy, reserved-prefix, hard-session-lease-bypass-inventory, resolve-web-provider-host, retirement/runtime-block/source-retirement/management-retirement/image-handler-retirement, migration-168, combo-auto-candidate-expansion, virtual-auto-combo, executor-map-golden and siblings), plus `typecheck:core`, `check-file-size`, and `check-changelog-integrity` clean. Thanks for the thorough provenance-hold retirement work — appreciated.
2026-08-28 06:52:46 -03:00

262 lines
14 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
---
title: "Stealth Guide"
version: 3.8.40
lastUpdated: 2026-06-28
---
# Stealth Guide
> **Source of truth:** `open-sse/utils/tlsClient.ts`, `open-sse/services/{claudeCodeCCH,claudeCodeFingerprint,claudeCodeObfuscation,claudeCodeCompatible}.ts`, `open-sse/config/cliFingerprints.ts`, `src/mitm/`
> **Last updated:** 2026-06-28 — v3.8.40
> **Audience:** Engineers maintaining provider-specific stealth integrations.
OmniRoute integrates with providers whose edges actively fingerprint non-official clients (TLS JA3/JA4, header ordering, JSON body shape, integrity tokens). This page documents the stealth surfaces OmniRoute exposes and where they are implemented.
## Legal and Ethical Notice
Stealth features exist so OmniRoute can act as a compatibility layer between user-owned official accounts (Claude Code CLI, Codex, Antigravity, Cursor, etc.) and OmniRoute's unified API. They are **not** for evading fraud detection, sharing credentials, or violating provider Terms of Service. The maintainers expect operators to comply with the upstream ToS they signed when creating accounts.
---
## TLS Fingerprinting Layer
### `open-sse/utils/tlsClient.ts` — wreq-js (Chrome 124)
Lazy-loaded `wreq-js` session that impersonates **Chrome 124 on macOS**. Used as a generic JA3/JA4 wrapper for upstreams behind Cloudflare. Falls back to native fetch when `wreq-js` is not installed (`available = false`).
- Singleton session: `browser: "chrome_124", os: "macos"`
- Proxy resolution (priority): `HTTPS_PROXY``HTTP_PROXY``ALL_PROXY` (also lower-case)
- Timeout: `TLS_CLIENT_TIMEOUT_MS` (inherits from `FETCH_TIMEOUT_MS`, default 600000)
- `wreq-js` Response is fetch-compatible (`headers`, `text()`, `json()`, `clone()`, `body`).
## Claude Code Stealth Bundle
When `cliCompatMode` is on, OmniRoute reshapes outgoing Claude requests so they are indistinguishable from `claude-cli` traffic. Three modules collaborate:
### `claudeCodeFingerprint.ts`
Computes the 3-char `cc_version` fingerprint embedded in the billing header:
```
SHA256(SALT + msg[4] + msg[7] + msg[20] + version)[:3]
```
- `FINGERPRINT_SALT = "59cf53e54c78"` (hardcoded; matches official client)
- Inputs: chars at index 4, 7, 20 of the first user message text + version string
- Output: 3-char hex prefix
### `claudeCodeCCH.ts` (Client Content Hash)
Server-side integrity check the official Claude Code CLI computes via Bun/Zig. OmniRoute reimplements with `xxhash-wasm`:
1. Serialize body with `cch=00000;` placeholder
2. `xxhash64(bytes, seed) & 0xFFFFF`
3. Zero-padded 5-char lowercase hex
4. Replace `cch=00000;` with the computed token
Constants:
- Seed: `0x6e52736ac806831e`
- Pattern: `/\bcch=([0-9a-f]{5});/`
### `claudeCodeObfuscation.ts`
Inserts a Unicode **zero-width joiner** (`U+200D`) after the first character of "sensitive" client names so upstream filters cannot grep them. Default word list:
```
opencode, open-code, cline, roo-cline, roo_cline, cursor, windsurf,
aider, continue.dev, copilot, avante, codecompanion
```
Applied to: `system` blocks, all `messages[].content`, and `tools[].description` / `tools[].function.description`. Operator-overridable via `setSensitiveWords()`.
### `claudeCodeCompatible.ts` — `anthropic-compatible-cc-*` providers
For third-party Anthropic relays that only accept "real Claude Code" traffic:
- `CLAUDE_CODE_COMPATIBLE_USER_AGENT = "claude-cli/2.1.219 (external, sdk-cli)"`
- `CLAUDE_CODE_COMPATIBLE_STAINLESS_PACKAGE_VERSION = "0.94.0"`
- `CLAUDE_CODE_COMPATIBLE_STAINLESS_RUNTIME_VERSION = "v26.3.0"`
- `anthropic-beta = "claude-code-20250219,interleaved-thinking-2025-05-14,effort-2025-11-24"` by default
- The per-connection "Enable redact-thinking beta" toggle adds `redact-thinking-2026-02-12` when a CC Compatible upstream specifically requires redacted thinking streams
- The per-connection "Enable summarized thinking display" toggle stores `providerSpecificData.requestDefaults.summarizeThinking` and adds `display: "summarized"` to CC Compatible thinking requests that did not already set a display mode
- `CONTEXT_1M_BETA_HEADER = "context-1m-2025-08-07"` (Opus/Sonnet 4.x family)
- Default path: `/v1/messages?beta=true`
Sister modules in the same bundle:
- `claudeCodeConstraints.ts` — temperature + cache-control rules
- `claudeCodeToolRemapper.ts` — tool-name remapping
- `claudeCodeExtraRemap.ts` — extra payload normalization
---
## Antigravity Stealth
Antigravity requests preserve caller text byte-for-byte. OmniRoute does not insert zero-width characters into prompts or rename/inject tools to imitate an IDE client.
### `antigravityHeaderScrub.ts`
Strips Stainless SDK markers (`x-stainless-lang`, `x-stainless-package-version`, `x-stainless-os`, `x-stainless-arch`, `x-stainless-runtime`, `x-stainless-runtime-version`, `x-stainless-timeout`, `x-stainless-retry-count`, `x-stainless-helper-method`) before forwarding.
### ⚠️ Risk: `ANTIGRAVITY_CREDITS=always` (account-ban hot spot)
`ANTIGRAVITY_CREDITS=always` (consumed by `open-sse/executors/antigravity.ts`) routes **every** request through Antigravity AI Credit Overages (paid Google credits) instead of letting Google's free-tier quota gate things. This is documented as a feature, but it is **the single most common ToS-violation report we see** — multiple Google Ultra accounts have been banned with `403 / "service disabled for ToS violation" / insufficient_quota` after running for a few hours with `=always`.
The upstream enforcement is on **Google's side**, not anything OmniRoute can prevent. The env var name and the existing docs make it sound like a safe knob to flip; it isn't.
**Why this draws abuse detection more aggressively than free-tier-only usage:**
- Sustained automated spend on a single Google account flags differently than free-tier hits-quota-and-stops.
- Credit overages have no rate ceiling, so a misconfigured client can burn through several hundred USD in minutes and look like API-key resale or bot traffic.
- Multiple OmniRoute users hitting overage credits in parallel from the same external IP compounds the signal.
**Recommended posture:**
1. Keep the default `ANTIGRAVITY_CREDITS=off` unless the operator explicitly accepts paid-credit and account-enforcement risk. `retry` sends the normal request first and injects credits at most once after an eligible quota 429; `always` injects credits on the first request.
2. **Spread load across providers via Auto-Combo** (`model: "auto"` or `kr/glm/etc`-combo) instead of saturating a single Antigravity account.
3. **Set per-connection RPM limits** in the Antigravity provider's edit page (Dashboard → Providers → Antigravity → connection → rate limit). 3060 RPM is a defensible upper bound for sustained use.
4. **Use stable, operator-controlled upstream networking** and avoid sharing one account across unrelated users or workloads.
5. **If banned**: appeal via `support.google.com` → "Restore Workspace/Account access" with the exact `quota_exceeded` / `service disabled` response body Google sent. Restoration is not guaranteed.
The environment reference documents the account and spend implications of each credits mode.
Touch points:
- `open-sse/executors/antigravity.ts` — reads `process.env.ANTIGRAVITY_CREDITS`
- `src/lib/oauth/providers/antigravity.ts` — credential plumbing
- Original incident report: Discussion [#1183](https://github.com/diegosouzapw/OmniRoute/discussions/1183)
---
## CLI Fingerprint Registry — `open-sse/config/cliFingerprints.ts`
Per-provider table that pins **exact** header ordering and JSON body field ordering captured from mitmproxy traces of the official CLIs. Currently registered: `codex`, `claude`, plus runtime-derived profiles in `providerHeaderProfiles.ts` for `antigravity` and `github`.
```ts
interface CliFingerprint {
headerOrder: string[]; // case-sensitive
bodyFieldOrder: string[]; // top-level JSON keys
userAgent?: string | (() => string);
extraHeaders?: Record<string, string>;
}
```
Toggle per provider via env (see below). When disabled, headers/body keys appear in whatever order Node/JSON gave them — easy to fingerprint.
---
## MITM Proxy (Antigravity, Linux/macOS/Windows)
For CLIs whose binaries cannot be redirected via `OPENAI_BASE_URL`, OmniRoute runs a local TLS-terminating proxy. Endpoints live under `src/app/api/cli-tools/antigravity-mitm/`.
| Method | Endpoint | Purpose |
| ------ | --------------------------------------- | ------------------------------------------------ |
| GET | `/api/cli-tools/antigravity-mitm` | Status — running, pid, dnsConfigured, certExists |
| POST | `/api/cli-tools/antigravity-mitm` | Start MITM (requires `apiKey` + `sudoPassword`) |
| DELETE | `/api/cli-tools/antigravity-mitm` | Stop MITM |
| GET | `/api/cli-tools/antigravity-mitm/alias` | List model aliases |
| PUT | `/api/cli-tools/antigravity-mitm/alias` | Save model aliases for a tool |
Target intercepted host: **`daily-cloudcode-pa.googleapis.com`** (Antigravity's upstream).
### Start sequence (`src/mitm/manager.ts::startMitm`)
1. Generate self-signed cert via `selfsigned` (RSA-2048, SHA-256, 1y) — `cert/generate.ts`
2. Install cert to system trust store — `cert/install.ts`
3. Add hosts entry `127.0.0.1 daily-cloudcode-pa.googleapis.com``dns/dnsConfig.ts`
4. Spawn `src/mitm/server.cjs` with `ROUTER_API_KEY` + `MITM_LOCAL_PORT` (default `443`)
5. Persist PID to `<DATA_DIR>/mitm/.mitm.pid`
### Linux dynamic trust-store detection — `cert/install.ts`
`getLinuxCertConfig()` walks a priority list and picks the first existing directory:
| Distro family | Directory | Update command |
| ------------------------ | ------------------------------------------- | ------------------------ |
| Debian / Ubuntu | `/usr/local/share/ca-certificates` | `update-ca-certificates` |
| Arch / CachyOS / Manjaro | `/etc/ca-certificates/trust-source/anchors` | `update-ca-trust` |
| Fedora / RHEL / CentOS | `/etc/pki/ca-trust/source/anchors` | `update-ca-trust` |
| openSUSE | `/etc/pki/trust/anchors` | `update-ca-certificates` |
Cert filename: `omniroute-mitm.crt`. Fingerprint match via `getCertFingerprint()` (SHA-1 of DER).
Additionally, `updateNssDatabases()` installs into per-user NSS DBs when `certutil` is available: `~/.pki/nssdb`, `~/snap/chromium/.../nssdb`, all Firefox profiles (including snap), under the nickname **`OmniRoute MITM Root CA`**.
### macOS / Windows
- **macOS:** `security add-trusted-cert -d -r trustRoot -k /Library/Keychains/System.keychain`
- **Windows:** elevated PowerShell → `certutil -addstore Root`
### Auth
All MITM endpoints require management auth (`requireCliToolsAuth`). The sudo password is cached in module scope (never `globalThis`) and cleared on `stopMitm()`.
---
## User-Agent Overrides — env vars (`.env.example` section 12)
| Variable | Default |
| ------------------------ | --------------------------------------------------------------- |
| `CLAUDE_USER_AGENT` | `claude-cli/2.1.219 (external, cli)` |
| `CODEX_USER_AGENT` | `codex-cli/0.142.0 (Windows 10.0.26200; x64)` |
| `GITHUB_USER_AGENT` | `GitHubCopilotChat/0.54.0` |
| `ANTIGRAVITY_USER_AGENT` | `antigravity/2.0.1 linux/arm64 google-api-nodejs-client/10.3.0` |
| `KIRO_USER_AGENT` | `AWS-SDK-JS/3.0.0 kiro-ide/1.0.0` |
| `QODER_USER_AGENT` | `Qoder-Cli` |
| `CURSOR_USER_AGENT` | `Cursor/3.4` |
Consumed by `open-sse/executors/base.ts::buildHeaders()` via dynamic lookup. **Bump these when providers release new CLI versions** — stale UA strings start getting rejected as outdated clients.
## CLI Compatibility Mode Toggles (`.env.example` section 13)
| Variable | Effect |
| -------------------------- | ------------------------------- |
| `CLI_COMPAT_CODEX=1` | Codex fingerprint |
| `CLI_COMPAT_CLAUDE=1` | claude-cli fingerprint |
| `CLI_COMPAT_GITHUB=1` | GitHub Copilot Chat fingerprint |
| `CLI_COMPAT_ANTIGRAVITY=1` | Antigravity fingerprint |
| `CLI_COMPAT_KIRO=1` | Kiro |
| `CLI_COMPAT_CURSOR=1` | Cursor |
| `CLI_COMPAT_KIMI_CODING=1` | Kimi Coding |
| `CLI_COMPAT_KILOCODE=1` | KiloCode |
| `CLI_COMPAT_CLINE=1` | Cline |
| `CLI_COMPAT_ALL=1` | Enable all of the above |
The provider IP is **always preserved** — the toggle only reshapes the request wire image, it does not switch IP egress.
---
## Inbound Header Sanitization
OmniRoute scrubs inbound client headers before forwarding so a request that arrives from Cursor doesn't leak `User-Agent: Cursor/X.Y.Z` to a Claude upstream. See `src/shared/constants/upstreamHeaders.ts` for the denylist, kept in lockstep with the Zod schemas and unit tests.
---
## Updating Fingerprints When a Provider Rotates
1. Capture official CLI traffic with `mitmproxy` (TLS interception + dump)
2. Extract JA3/JA4 and the literal header order
3. Update the relevant `CLI_FINGERPRINTS[...]` entry
4. Bump matching `*_USER_AGENT` default in `.env.example`
5. If the TLS handshake itself changed, update the relevant provider wrapper or the wreq-js `browser:` option
6. Run the provider-specific TLS tests and a manual canary against the live provider
7. Ship in a patch release; document in `CHANGELOG.md`
---
## Tests
- `open-sse/services/__tests__/claudeTlsClient.test.ts` — shared TLS wrapper behavior
- `tests/unit/anthropic-cache-fingerprint.test.ts` — fingerprint determinism
- `tests/unit/chatgpt-web-source-retirement.test.ts` — common ChatGPT Web stealth source remains absent while Codex Web stays present
---
## See Also
- [RESILIENCE_GUIDE.md](../architecture/RESILIENCE_GUIDE.md) — what happens when a stealth path gets a `403`
- [TROUBLESHOOTING.md](../guides/TROUBLESHOOTING.md)
- [ENVIRONMENT.md](../reference/ENVIRONMENT.md) — full env reference
- [CLI-TOOLS.md](../reference/CLI-TOOLS.md) — operator view of the MITM workflow