docs: m17 hang residual diagnosis (model, not engine abort) - #20
Conversation
Record evidence that seal hang after /v1 is Spark Ollama gpt-oss:20b non-response; secondary slash-in-version plugin install hang is harness handoff only. No agent seal claim.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 129d49623b
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| return chat completions in time. Operator mitigations: smaller model | ||
| (`VERDICT_LLM_MODEL=llama3.1:8b`), free Ollama capacity, or accept fallback PASS | ||
| (not agent seal). |
There was a problem hiding this comment.
For the model non-response scenario described here (0 bytes from chat completions and the agent process staying hung), operators cannot “accept fallback PASS”: caseforge investigate only reaches runLocalEvtxAutoFallback after the verdict run child exits and verification continues (packages/caseforge-cli/src/commands/investigate.ts:474-482), while the smoke script treats timeout exits 124/137 as immediate FAIL before the fallback PASS path (scripts/spark-local-seal-smoke.sh:238-250). This line will mislead anyone trying to mitigate the m17 hang with the documented smoke flow, because the fallback only helps fail-fast/verify-fail cases, not a stuck LLM stream.
Useful? React with 👍 / 👎.
1 participant
Summary
/v1+ PR test(smoke): clarify agent hang timeout after /v1 fix #19 hang messaging.gpt-oss:20bchat completions hang (0 bytes / 30s curl); doctor/v1and/api/tagsstill OK.llama3.1:8bcompletes.0.0.0-agent/m4-…contains/→ background@opencode-ai/plugin@${version}→git ls-remoteGitHub hang. Fix in verdict-opencode + rebuild; avoid dual-edit with opencode lane.Test plan
npm run buildnode scripts/selftest.mjs(120 passed)