feat(runner): one-shot retry on opencode null-with-no-errorKind + retry chip visibility#85
Closed
chorus-codes wants to merge 1 commit into
Closed
feat(runner): one-shot retry on opencode null-with-no-errorKind + retry chip visibility#85chorus-codes wants to merge 1 commit into
chorus-codes wants to merge 1 commit into
Conversation
Victor caught this on the PR #83 audit: qwen3.6-plus reviewer ran on opencode-go, exited cleanly with empty output (no errorKind, no message), and went straight to fallback chain advance — no retry attempt. The PR #79 retry classifier returned false because isRetryableErrorKind(undefined) was hard-false. That's the right default for codex/claude/gemini (a null with no kind usually means the model genuinely produced nothing — retry would produce nothing again), but opencode-go's gateway has known transport flakes where a second attempt succeeds with the same prompt. Fix: extend isRetryableErrorKind to accept an optional `lineage` hint. When `kind` is undefined AND `lineage === 'opencode'`, treat as retryable. Other lineages keep the conservative default. The lineage hint does NOT override an explicit non-retryable kind — auth/quota/ db-corrupt are still terminal regardless of lineage. Both reviewer-driver and doer-driver call sites now pass `entry.lineage` so the chain step picks up the new behaviour. Retry visibility: no UI work needed — the existing `transient_retry` cli_warning already renders as an amber chip on the participant card via participant-card.tsx's `participant.warnings` block. The chip appears the moment retry fires; the message reads "Transient X failure on Y/Z — retrying once before advancing fallback." Tests: 974 -> 975 passing (+5 new cases on the lineage hint behavior).
Owner
Author
3 tasks
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.
Summary
Victor caught this on the PR #83 audit screenshot. The opencode-go qwen3.6-plus reviewer exited with an empty answer (no `errorKind` classified) and went straight to claude-sonnet-4-6 fallback — the PR #79 retry never fired because `isRetryableErrorKind(undefined)` is hard-false.
Fix
Extended classifier signature to `isRetryableErrorKind(kind, lineage?)`. The new rule: `kind === undefined && lineage === 'opencode'` → retry once. Other lineages unchanged.
Why opencode specifically:
Retry visibility (no UI work needed)
The `transient_retry` cli_warning already renders as an amber chip on the participant card via the existing `participant.warnings` block in `participant-card.tsx`. With this PR, opencode failures that previously went silent-to-fallback now fire the warning, so users see "TRANSIENT_RETRY — Transient failure on opencode/opencode-go/qwen3.6-plus — retrying once before advancing fallback." before the swap banner appears.
Tests
974 → 975 passing. New tests cover:
Test plan