METR/hawk

design-review

Review of a pull request, written as one top-level comment. Judges the design (whether the problem is worth solving, whether this is the simplest approach that fits the monorepo, which existing mechanism or alternative it passed over, what complexity could be…

原文语言：英语

debug-stuck-eval

Debug stuck Hawk/Inspect AI evaluations. Use when user mentions "stuck eval", "eval not progressing", "eval hanging", "samples not completing", "eval set frozen", "runner stuck", "500 errors in eval", "retry loop", "eval timeout", or asks why an evaluation…

原文语言：英语

database-migrations

How to create and use our alembic database migration tool. Use when making changes to models.py.

原文语言：英语

babysit-eval

Babysit a running Hawk eval over the ACP control channel — watch samples live, approve/reject parked tool calls, steer or cancel samples. Use when the user asks to "babysit" an eval, approve its tool calls, watch a sample live, answer a pending approval, or…

原文语言：英语

full-stack-dev

How to develop the frontend and backend together. When you want to make changes to the UI, use this.

原文语言：英语

view-results

View and analyze Hawk evaluation results. Use when the user wants to see eval-set results, check evaluation status, list samples, view transcripts, or analyze agent behavior from a completed evaluation run.

原文语言：英语

sync-docs

Use when the user asks to update, sync, or refresh the Hawk docs (quickstart, configuration reference, infrastructure description and diagrams) to reflect changes since the last documentation sync. Triggers on "update the docs", "sync docs", "docs are stale",…

原文语言：英语

monitoring

Monitor Hawk job status, view logs, and diagnose issues. Use when the user wants to check job progress, view error logs, debug a failing job, or generate a monitoring report for a Hawk evaluation run.

原文语言：英语

smoke-tests

Run smoke tests on a deployed environment to ensure basic functionality.

原文语言：无法判定

auditing-dependency-bumps

Use when reviewing a Dependabot (or similar) dependency-bump PR — npm/pnpm minor/patch group bumps or GitHub Actions SHA bumps — and you need to do a supply-chain audit before approving and merging. Covers downloading and inspecting package contents, checking…

原文语言：英语

managing-dlqs

Use when an SQS dead-letter queue has accumulated messages and you need to investigate, redrive, or drain them — e.g. a DLQ-depth CloudWatch alarm fires, the importer is failing, or you're cleaning up after a known bug fix has shipped.

原文语言：英语