dev-workflow

Autonomous implementation worker for the dev-workflow pipeline. Wraps the start-development skill in an isolated subagent context. Dispatched by full-cycle / epic for the non-interactive development stage. Branches, implements with TDD, opens one PR per repo, and returns a flat key/value result. Use via subagent_type from an orchestrator — not for ad-hoc edits.

Autonomous rework worker for the dev-workflow pipeline. Wraps the address-pr-comments skill in an isolated subagent context. Dispatched by full-cycle / epic inside the review loop and test loop to implement requested changes on an existing PR's branch and reply to the review. Use via subagent_type from an orchestrator.

Per-task pipeline driver for the epic skill. Wraps full-cycle in an isolated subagent context so an epic can run each task end to end while keeping each task's work out of the epic orchestrator's context. CRUCIALLY retains the Agent tool (no `tools` restriction) so it can itself dispatch the per-stage workers (developer / reviewer / tester / fixer / spec-writer / pr-state-reader) as nested subagents — requires Claude Code v2.1.172+ for nesting. Use via subagent_type from the epic scheduler.

Read-only state probe for the dev-workflow pipeline. Used by full-cycle for entry/resume detection, resolving PR numbers linked to a story, and reading the authoritative review/test decision from GitHub, keeping raw PM/GitHub JSON out of the orchestrator context. It reads and returns exactly one line — it never edits, branches, or reviews. Use via subagent_type from an orchestrator.

Autonomous code-review worker for the dev-workflow pipeline. Wraps the review-pr skill in an isolated subagent context. Dispatched by full-cycle / epic for the review stage. Triggers a fresh dev build CI on current HEAD, runs the multi-perspective review, submits a formal GitHub review, and returns a flat key/value result. Use via subagent_type from an orchestrator.

Address PR review feedback in the current session — reads new comments, file-level inline comments, and review decisions since the last commit, then implements the required changes and replies to the PR with a summary of fixes. Use when a user asks to address PR comments, respond to review feedback, fix review notes, or iterate on a PR.

Use when a user wants to capture a feature idea as a formal story, create a ticket, write to the backlog, or says 'create a story', 'write a ticket', 'add to backlog', or describes a feature they want tracked.

Use when a large initiative must be decomposed into many coordinated tasks and driven to completion, when a user says 'epic', 'break this down', 'orchestrate this feature', 'run this whole initiative', or '/start epic', or provides a high-level multi-task feature summary that spans more than one PR or repo. Resumable: re-invoke against an existing epic slug to continue an interrupted run.

Drives the entire dev-workflow lifecycle end to end in order — create-story → write-spec → start-development → review-pr → test-pr — looping back through review and testing until both pass. Use when a user wants to run the whole development pipeline for a feature, take a story 'all the way through', or says 'full cycle', 'end to end', 'run the whole workflow', or '/start full-cycle'. Resumable: re-invoke at any point and it detects current story/PR state and enters at the correct stage.

Comprehensive multi-perspective PR review (Product Manager, Developer, QA, and Architect lenses) that compares implementation against story requirements and CI/CD results. Includes automatic first-review vs re-review mode detection. Always use this when a user asks to review a PR, check a pull request, or validate an implementation against requirements.

2 hooks across 2 events