Move provider model registries, stream wrappers, OAuth helpers, and LLM utilities into src/llm with plugin-sdk barrels instead of depending on the old embedded runtime layout.
Fix runtime context placement so hidden runtime context is model-visible before the active user turn without persisting as a visible/session message.
Verification:
- git diff --check origin/main...origin/pr/86995-merge
- gh pr checks 86995 --repo openclaw/openclaw --watch=false
- gh run rerun 26493979156 --repo openclaw/openclaw --failed
- gh run watch 26493979156 --repo openclaw/openclaw --exit-status
- CodeQL run 26493979156 attempt 2, Security High (mcp-process-tool-boundary) job 78066719467 passed
Validates forced plugin harness support for the requested provider/model before pinning Codex or any other plugin harness. This prevents an explicitly forced Codex runtime from accepting unsupported OpenAI-like providers through a hardcoded bypass while preserving implicit PI fallback and CLI runtime alias passthrough.
Regression coverage covers forced Codex rejection for unsupported openai/openai-codex support, Codex provider support declarations, CLI attempt routing, pi-embedded auth/profile forwarding fakes, Testbox scenario probes, and live Docker Codex plugin E2E.
Thanks @cathrynlavery.
Keep the Codex app-server full attempt watchdog armed after a terminal turn notification is queued, so a wedged notification projector cannot leave a run stuck indefinitely.
Proof:
- `git diff --check origin/main...HEAD`
- `node scripts/run-oxlint.mjs extensions/codex/src/app-server/run-attempt.ts extensions/codex/src/app-server/run-attempt.test.ts`
- `node scripts/run-vitest.mjs run extensions/codex/src/app-server/run-attempt.test.ts --testNamePattern "keeps the attempt watchdog armed"` passed in PR proof (`1 passed | 232 skipped`)
- `OPENCLAW_TESTBOX=1 pnpm check:changed` passed in `tbx_01kskyg44ej461k574jee8ffjc`
- CI required checks green after `build-artifacts` rerun job `78031279635` passed
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
Regression test for the binary stall fix: when rawResponseItem/completed
arrives with a non-assistant type (e.g. "reasoning") and all tracked
items have completed, the completion idle watch must stay armed so the
stall is caught in 60s, not 30 minutes.
Refs openclaw/openclaw#87071
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When the codex binary emits rawResponseItem/completed and all tracked
items have completed (activeTurnItemIds empty, no active requests), the
binary should deliver turn/completed imminently. Previously, a
rawResponseItem/completed that didn't qualify as a post-tool assistant
completion would actively disarm the completion idle watch, leaving only
the 30-minute terminal timeout to catch a stalled binary. This caused
turns to hang for up to 30 minutes when the OpenAI Responses API fails
to deliver response.completed to the binary.
Now, rawResponseItem/completed with no active items arms the 60s
completion idle watch and is excluded from the disarm path, so stalled
binaries are detected in 60s instead of 30 minutes.
Refs openclaw/openclaw#87071
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Project newer external OpenClaw chat history into resumed Codex app-server threads when the saved binding is older than user-visible transcript messages, while filtering Codex-owned mirror records on consecutive resumes.
Thanks @TurboTheTurtle!
Keep Codex app-server turn timeouts within the Codex runtime boundary so they interrupt the active turn without retiring the shared app-server client, poisoning auth-profile cooldowns, or falling through to generic provider/model fallback.
Preserve concrete non-timeout provider failures for auth-profile rotation and fallback, and add regression coverage for prompt-stage timeouts, assistant idle timeouts, auth-profile cooldowns, and app-server timeout handling.
Thanks @pashpashpash.
