fix(runner): mcp_handshake_failed is transient, not auth#89
Merged
Conversation
Codex's bundled MCP server boots racily — slow first start, handshake timeout under load, server crash on first start. The error LOOKS auth-shaped (was originally classified terminal under "auth issue, same remediation as token_refresh_lost") but real codex auth failures always surface as token_refresh_lost. mcp_handshake_failed is almost always a transient boot race. Caught when codex hit this on the PR #87 audit chat — the run shows "MCP_HANDSHAKE_FAILED — failed: handshaking with MCP server failed" and went straight to claude-sonnet-4-6 fallback with no retry. Victor confirmed codex was working fine in the previous chat just minutes earlier — definitionally transient. Fix: move mcp_handshake_failed from terminal to transient in isRetryableErrorKind. Worst case: wastes one extra codex call on a genuine misconfig (fails twice and advances to fallback as before). Best case: catches the boot race and saves the fallback swap. Note: kindToStatus still maps mcp_handshake_failed -> auth_invalid for cli-health tracking. That's a separate concern (it triggers the 10-min precheck cooldown + "Auth broken" badge); flagging as follow-up but PR #81's auto-heal on cred-file mtime change covers the common recovery path. Worth revisiting in a dedicated PR if the badge proves sticky in practice. Tests: 978 passing. One terminal test removed, one transient test added.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
Codex hit `MCP_HANDSHAKE_FAILED` on the PR #87 audit chat, went straight to claude-sonnet-4-6 fallback with no retry attempt. Victor confirmed codex was working fine minutes earlier — definitionally transient.
Root cause
`mcp_handshake_failed` was classified terminal in `isRetryableErrorKind` under the assumption "auth issue, same remediation as token_refresh_lost." That assumption is wrong: real codex auth failures always surface as `token_refresh_lost` (the explicit refresh-token error). `mcp_handshake_failed` is almost always codex's bundled MCP server booting racily.
Fix
One-line: add `mcp_handshake_failed` to the transient case list. Worst case wastes one extra codex call on a genuine misconfig (fails twice, advances to fallback as before). Best case catches the boot race.
Out of scope (follow-up)
`kindToStatus` in `cli-health.ts` still maps `mcp_handshake_failed` → `auth_invalid`, which triggers the 10-min precheck cooldown and "Auth broken" badge. PR #81's mtime auto-heal covers the common recovery path (next `codex login` or any cred file touch clears it). Worth revisiting separately if the badge proves sticky.
Tests
978 passing. One terminal test removed (`mcp_handshake_failed is terminal`), one transient test added.
Test plan