workflow

Plugin workflow from marketplace trongio in trongio/claude-workflow. Commands read from the repo's manifest — 27 days ago.

Skills (5)

• board

  Show the task board - start the live board server (milestone progress, ETA, blocked cards, waiting on owner) or print the board stats in the terminal. Invoke as /workflow:board [stats|serve].

• gates

  Run every quality gate configured in .claude/workflow.json for the current change set, in parallel, and report a compact pass/fail table with only the failing output. Use before every review, commit, and at the end of every task.

• init

  Bootstrap the current repo for the workflow - create tasks/ with its README, detect the stack and write .claude/workflow.json with prefilled gates, and add a Workflow section to CLAUDE.md. Safe to re-run. Invoke as /workflow:init.

• next

  Pick up work from the board. Claims the next cards (two in parallel by default, WIP limit from workflow.json), runs each through /workflow:task concurrently, and keeps pulling cards as they finish. Invoke as /workflow:next, /workflow:next 1, or /workflow:next drain.

• task

  Run one task through the manager loop (claim, plan, optional mocks, delegate, integrate, gates, simplify, risk-scaled review, verify, UX critique, record, merge). Moves the card on the board at every stage. Invoke as /workflow:task <card ID | description> [--headless].

Subagents (8)

• implementer

  Implements one bounded workstream end to end (code, tests, gates) on the branch the brief names and reports a diff summary. Use for any feature or fix beyond a trivial one-file edit.

• mechanic

  Routine mechanical work with no design judgment - run the gates and report, apply formatter/linter fixes, card moves and `## Outcome` entries, mechanical renames, translation or fixture entries, exactly as briefed.

• planner

  Plans one task before any code is written. Reads the card, the project docs and the relevant code, then returns a file-level plan with acceptance criteria, risks, a risk level, and a self-contained brief per workstream. Read-only.

• researcher

  Answers one technical question that needs library internals, docs, external APIs or runtime behavior, with a sourced answer and code references. Read-only.

• reviewer

  Adversarial code review of a task's diff. Hunts correctness bugs, security holes, missing tests, convention breaks and needless complexity. One run for standard risk, two independent runs for high risk. Read-only.

• scout

  Fast read-only lookups - where a symbol, route or component lives, which files reference X, a short summary of one file or doc section. Use when the answer is a path or a few lines, not an analysis.

• ux-critic

  Judges a finished UI change as a demanding product designer would - is it clear, does it look intentional rather than templated, does it help the user finish the job, does it hold up on mobile. Grades screenshots (the verifier's, or its own) and returns ranked fixes. Read-only.

• verifier

  Drives the running dev app in a real browser (Playwright MCP or a project check script) to exercise a task's user-facing surface end to end at desktop and mobile widths, and reports what actually rendered. Runs once per task after the reviews.

Similar plugins