claude/skills/feature-workflow/SKILL.md
End-to-end feature development: brainstorm -> plan -> implement -> review -> ship. Use when the user invokes /feature-workflow or asks to run the full feature development pipeline.
npx skillsauth add dragonkid/dotfiles feature-workflowInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
You are orchestrating a complete feature development pipeline. This file is the control flow — each phase's detailed instructions live in references/. You MUST execute phases in order and MUST NOT silently skip any phase.
Feature request: $ARGUMENTS
Branch setup comes first so that all artifacts (design docs, plans, code) live on the same branch. The branch name is derived from $ARGUMENTS.
Smart skip analysis: Check if the user is already on a feature branch or worktree. If so, recommend skipping.
feat/ or fix/ branch, or in a worktree? -> Recommend: "You're already on branch X. Skip branch setup?"If proceeding: Read references/phase-1-branch.md and follow it completely.
Update TodoWrite: phase=1-complete, branch_mode=<worktree|branch|skipped>, base_branch=<name>.
Read references/phase-2-design.md and follow it completely.
GATE 2 — When Phase 2 completes, use AskUserQuestion:
| Choice | Action | |--------|--------| | Start implementation | Record execution mode in TodoWrite. Proceed to Phase 3. | | Revise plan | Re-invoke writing-plans skill, then return to this gate. | | Stop workflow | End. |
Update TodoWrite: phase=2-complete, execution_mode=<choice from design phase>.
Read references/phase-3-implement.md and follow it completely.
GATE 3 — Use AskUserQuestion:
| Choice | Action | |--------|--------| | Proceed to review | Proceed to Phase 4. | | Continue implementing | Stay in Phase 3, continue work, then return to this gate. | | Stop workflow | End. |
Update TodoWrite: phase=3-complete.
ENTRY GATE — Before doing anything, use AskUserQuestion to let the user choose the review level. Provide change scope context (files changed, lines added) to help them decide:
| Choice | Action |
|--------|--------|
| Full review | MANDATORY: Read references/phase-4-review.md FIRST, then follow all 4 steps exactly. The reference file contains the agent matrix table — you MUST use it to determine which agents to launch. Do NOT improvise the agent list from memory. The matrix includes always-on agents (verification, security, code review, simplify) plus conditional agents (language-specific, architecture, test coverage, codex). |
| Lightweight | Run only the verification loop (build, types, lint, tests) and skip the multi-agent dispatch. |
| Skip review | Proceed directly to Gate 4. |
Why full review matters: A single reviewer catches surface issues, but cross-model multi-agent review catches deeper problems — the FOMC multi-day parsing bug (HIGH) was found by a Python reviewer, the cache deadlock was found by a code reviewer, and the fan-out concurrency risk was flagged independently by 3 different agents. Single-agent review would have missed most of these.
GATE 4 CHECKPOINT — Before presenting the Gate 4 question, verify:
If any item is unchecked, go back and complete it before proceeding.
GATE 4 — Use AskUserQuestion:
| Choice | Action | |--------|--------| | Continue to ship | Proceed to Phase 5. | | Fix issues first | Address issues, re-run affected agents, then return to GATE 4 CHECKPOINT (not Gate 4 directly — the checkpoint re-verifies all 4 steps including docs updates). | | Stop workflow | End. |
Update TodoWrite: phase=4-complete.
Read references/phase-5-ship.md and follow it completely.
Announce: "Feature workflow complete."
tools
Use when user wants to manage TODO items - adding tasks, listing pending items, marking done, removing, or searching. Triggers on /todo command or TODO-related requests.
tools
Use when creating a new skill, updating an existing skill, or troubleshooting skill frontmatter, gating, or slash command registration issues in OpenClaw.
tools
手动触发自我改进与记忆维护。分析近期 memory 文件,维护工作区文件,提出改进提案。触发方式:/self_improve 或用户说"自我改进"、"self improve"。
testing
测试 OpenClaw 配置中 LLM provider 的可用性。触发方式:/provider_check 或用户问"测试 provider 可用性"、"检查 provider 状态"、"provider 健康检查"。