Quoting Geoffrey Huntley
- ID
- 34801
- Status
- summarized
- Published
- 13 Oct 2026, 9:32 PM
- Fetched
- 13 Oct 2026, 9:44 PM
- Provider
- Simon Willison
- Category
- developer-ai
- Original URL
- https://simonwillison.net/2026/Oct/13/geoffrey-huntley/
- Source URL
- https://simonwillison.net/atom/everything/
Summary
- Score
- 7.0
- Created
- 13 Oct 2026, 9:44 PM
- Tags
- Audience
- developersvibe_codersai_agent_users
What happened
Simon Willison quotes Geoffrey Huntley describing a fix for coding agents that ignore project diagnostics tooling: when the agent bypasses the diagnostics module and calls observability modules directly, Huntley writes a pre-commit hook that inspects the git diff, fails the commit, and injects a prompt telling the agent to use the diagnostics library instead. Huntley says relying on updating AGENTS.md alone leaves a delta that drifts, whereas the hook approach made the agent 'ten out of ten'. The post is a short quotation, not a benchmark or write-up, so no measurements, versions, or repo are given.
Why it matters
If you run coding agents in a repo with a required internal library (logging, diagnostics, auth, API client), this argues for enforcing it in a pre-commit hook that fails the commit and feeds a corrective prompt back to the agent, rather than trusting AGENTS.md instructions that drift over a long session. The concrete decision for your team is where to put that hook and what string to inject — the quote claims the enforcement-plus-prompt loop is what changed the agent's behaviour, not more documentation. There is no Malaysian or SEA angle in this text; the value is the pattern, not local news.
Discussion angle
Ask the room which of their project rules are currently 'documented in AGENTS.md' versus 'enforced by a hook that fails the commit and re-prompts the agent' — and what the injected prompt text should actually say to make the agent comply instead of retrying the same bypass.