Greptile Summary

This PR appends a five-point ### Verification checklist to OPENAI_GPT5_EXECUTION_BIAS in extensions/openai/prompt-overlay.ts, adding checks for correctness, grounding, coverage, formatting, and safety before GPT-5 finalizes a response.

• The Safety bullet ("confirm scope before executing" for side-effecting actions) directly contradicts the pre-existing "Do not pause to summarize or verify before taking the next action" instruction in the same constant. Most consequential tool calls (file writes, API calls, shell commands) have side effects, so this effectively reintroduces a pause-and-confirm signal for the exact actions the execution bias was designed to suppress — likely causing inconsistent model behavior.

Confidence Score: 3/5

• Not safe to merge as-is; the Safety bullet contradicts the core execution bias directive.
• There is a clear P1 conflict: the new Safety verification item instructs the model to pause and confirm scope before side-effecting tool calls, while the surrounding prompt text explicitly says not to pause before taking the next action. This internal contradiction within the same exported constant will produce unpredictable model behavior across all GPT-5 sessions that use this overlay.
• extensions/openai/prompt-overlay.ts — the Safety bullet in the new Verification section

This is a comment left during a code review.
Path: extensions/openai/prompt-overlay.ts
Line: 72-80

Comment:
**Safety bullet contradicts execution bias principle**

The existing instruction on line 72 explicitly says "Do not pause to summarize or verify before taking the next action," but the new `Safety` bullet says "confirm scope before executing" for side-effecting actions (file writes, commands, API calls). Nearly every consequential tool call has side effects, so this tells the model to pause and confirm before the exact actions the constant is designed to push through without interruption. The two directives are in direct conflict within the same exported string, and a model reading the full text gets contradictory signals about whether to act or ask first.

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "feat(openai): add verification checklist..." | Re-trigger Greptile}

Superseded by consolidated final-sprint PRs. This PR's changes are incorporated into:

• feat(openai,agents): GPT-5.4 prompt discipline + personality bridge + injection scanning #67512 (OpenAI prompt stack + injection scanning)
• feat(google): add Gemini execution guidance via prompt overlay [v3 final 2/3] #67513 (Google Gemini guidance)
• feat(agents): task system parity — cancelled status, merge mode, activeForm, plan hydration [Phase 3.A] #67514 (Task-system parity bundle)

The v3 PRs were split across 6 separate PRs which made review harder and touched the same files redundantly. The final-sprint consolidation combines them into 3 clean, independently-mergeable PRs rebased on current main.