dev-tasks

Deep codebase exploration agent. Traces data flow, finds patterns, maps dependencies. Read-only.

Analyze git diff and update relevant docs and .claude/rules/. Makes targeted edits. No external system access.

Run Playwright E2E tests, parse results, suggest new test cases. Restricted to test-execution surface — no external systems beyond Playwright.

Post-implementation code review agent. Runs the 10-point checklist from /self-review. Never edits files. Never accesses external systems.

Sweep a product's open versions and surface promote-ready candidates, stale Planned versions, orphan tasks, and version-state-machine drift. Read-only — suggests fixes but doesn't apply them.

Orchestrator-side merge-polling loop for any PR the main session didn't push itself — subagent-produced PRs (most common), multi-agent fan-out (parallel subagents), or recovery from a prior session. Single-agent main-session work uses /ship-pr Phase 6.6 autonomous merge instead.

Create a new Monday.com task with duplicate checking and Ready-to-Start gate enforcement

Reuse-before-invent checklist for any UI-touching change. Forces a grep-and-survey of existing components, layouts, tokens, and patterns before writing new ones. Cuts visual drift and the "we already had a component for that" cycle.

Audit a consumer's dev-tasks setup. Verifies project-config is valid, MONDAY_API_KEY resolves, People-board lookup works for the current whoami, required Monday boards are reachable, and policy hooks are in place. Run after first install or when something feels off.

25 hooks across 7 events