Dev builds triage skill to tame GPT-6.2's overly paranoid code review
PawelHuryn · x · 2026-11-21
Pawel Huryn shares that GPT-6.2 is an excellent code reviewer but flags nearly everything as Medium or High severity. His fix: an independent-review skill that interrogates each finding — who hits it and how often, the harm in one sentence, and whether it crosses a red line (other users' data, lost work, unapproved irreversible writes).
On one build, nine review rounds produced noise; round ten with triage yielded 11 findings, none above Low, closing the stage with zero fixes needed.
The skill is part of his open-source product-engineering repo: 45 agent skills covering decide → build (Claude Code/Codex) → second-model review → release. MIT licensed, 687 stars.
More from coding & agent
- Mnemos 2.0: agent memory designed to preserve what matters for the next session — RileyRalmuto · 2026-11-21
- InferenceBench ranks agents on LLM serving speedup: Claude Opus 5.5 tops at 12.08x, agents lose to simple hyperparameter search — maksym_andr · 2026-11-21
- 4 architecture patterns to tame multi-agent system reliability, from a 27-year reliability engineer — bibryam · 2026-11-21
- AutoCut + FrameAtlas: open-source local-first video editing driven by Claude Code via MCP — moheshmohan · 2026-11-21
- InfoQ: What an agent harness is, and two ways to build one — bibryam · 2026-11-21
- The hidden 70 minutes: a Reddit cost breakdown shows agent workflows can cost more human time than manual — AlarmingPepper9193 · 2026-11-21