Explore

Rules for agent-written UI: semantics before ARIA, keyboard parity, visible focus, and contrast that actually passes

Pre-action checks for destructive ops, budget/turn limits, confirmation gates, and a structured audit log

Review an HTTP or RPC interface before it ships: naming, errors, pagination, versioning, and the parts that become permanent

Turns a git log range into a grouped CHANGELOG entry with breaking changes called out and links to issues/PRs

Conventional Commits, branch naming, and a PR description template as standing rules

Keep a long-running agent coherent past its context window: what to keep, what to summarize, and how to resume

Profile a messy dataset before analyzing it: types, missingness, duplicates, outliers, and the questions it cannot answer

Review a schema migration before it runs: locks, blast radius, deploy compatibility, and a rehearsed rollback

Audits outdated dependencies, upgrades in risk order, runs tests after each, and opens one PR per major bump

Build a real eval suite for an agent or prompt: cases, graders, a baseline, and a regression gate

Find why a test is flaky rather than retrying it: classify the cause, prove it, fix it, and quarantine honestly

Approval gates for risky agent actions: what needs asking, how to ask well, and what an approval does not cover

Triage, mitigate, communicate, and postmortem for production incidents, with timers and escalation rules

Turn a wall of logs into a ranked set of hypotheses, each with the evidence for it and the cheapest way to test it

Find why something is slow with measurement rather than intuition: baseline, profile, fix the top cost, verify

Turn an incident into a written postmortem: timeline from evidence, contributing factors, and actions that are actually assigned

Structured PR review with risk classification, a correctness/security/test/perf checklist, and severity-tagged output

Rules for agent-written Python: explicit failures, boring data structures, dependency restraint, and honest type hints

Map an unfamiliar codebase into an architecture summary, entrypoints, conventions, and run/test instructions

Turn a question into a sourced brief — search plan, graded sources, claim ledger, and a brief template

CLAUDE.md-style rules for secrets, input validation, authz, dependency hygiene, and PII logging

Read an EXPLAIN plan, find the actual bottleneck, and propose the smallest index or rewrite that fixes it

Rules for agent-written SQL and migrations: no unbounded writes, no blocking locks, always a rollback path

Red/green loop that picks an untested module, writes a failing test, implements it, and iterates with coverage deltas