Skip to content

fix(ship): give chatgpt-review plan-mode reviews pass-aware context - #620

Merged
BorisTyshkevich merged 1 commit into
mainfrom
fix/chatgpt-review-plan-revision-pass-context
Aug 6, 2026
Merged

fix(ship): give chatgpt-review plan-mode reviews pass-aware context#620
BorisTyshkevich merged 1 commit into
mainfrom
fix/chatgpt-review-plan-revision-pass-context

Conversation

@BorisTyshkevich

Copy link
Copy Markdown
Collaborator

Summary

  • mode === 'plan' in skills/chatgpt-review/scripts/lib/prompt.mjs built an identical review prompt on every pass, with no pass-aware branching — unlike mode === 'pr', which already tells ChatGPT to reassess its own earlier findings on later passes.
  • Observed live during a /ship 585.1 plan-review-loop run: ChatGPT re-raised two findings on pass 5 that were already rejected with cited repository evidence in the plan's own ## Review responses section on passes 1–3, burning review passes on re-litigation instead of surfacing new issues.
  • Pass 2+ now tells ChatGPT it is continuing the same conversation over the same plan, points it at ## Review responses for previously rejected findings, and asks it to explicitly refute the cited evidence rather than restate the original claim — mirroring the existing pr-mode pattern.

Test plan

  • node --test tests/*.test.mjs in skills/chatgpt-review/ — 43/43 pass, including new pass-1/pass-2 prompt-content assertions for mode: 'plan'.
  • skills/ is standalone agent tooling, not part of the main product's npm test/tsc/coverage gate (verified against root package.json scripts) — no product-code impact.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LwFPT465eDJqYcRa8HGNLz

mode 'plan' built an identical review prompt on every pass, unlike mode
'pr' which already tells ChatGPT to reassess its own earlier findings
on later passes. During a live plan-review-loop run (#585 phase 1),
ChatGPT re-raised two findings on pass 5 that were already rejected
with cited evidence in the plan's own "## Review responses" section
on passes 1-3, burning review passes on re-litigation instead of new
issues.

Pass 2+ now tells ChatGPT it is continuing the same conversation over
the same plan, points it at "## Review responses" for previously
rejected findings, and asks it to refute the cited evidence rather
than restate the original claim.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwFPT465eDJqYcRa8HGNLz
@BorisTyshkevich
BorisTyshkevich merged commit 44c2d22 into main Aug 6, 2026
8 checks passed
BorisTyshkevich added a commit that referenced this pull request Aug 6, 2026
Picks up PR #620 (chatgpt-review plan-mode fix, unrelated) so the
integration branch tracks main's tip before opening the Phase 1 PR.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwFPT465eDJqYcRa8HGNLz
@BorisTyshkevich
BorisTyshkevich deleted the fix/chatgpt-review-plan-revision-pass-context branch August 6, 2026 15:28
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant