菜单
Run the full verify suite and report results — no fixes applied. Use to check baseline health before starting work, after a merge, or when CI is red.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Create well-structured atomic commits with conventional commit messages. Use when ready to commit working changes.
Write a feature spec from requirements in .claude/input/. Use when starting a new feature, before implementation begins.
Orchestrate the full spec → implement → review → fix → commit pipeline. Use when shipping a complete feature end-to-end.
Break a concept document, draft, or product brief into independent PRDs that each feed into /ship. Use when input contains multiple features or a big-picture vision.
Ship decomposed PRDs in parallel. Analyzes file conflicts, groups into batches, runs /ship in isolated worktrees, merges results. Use after /decompose.
Implement a feature from a spec file in .claude/specs/. Use after a spec is written and approved.
| name | test |
| description | Run the full verify suite and report results — no fixes applied. Use to check baseline health before starting work, after a merge, or when CI is red. |
| disable-model-invocation | true |
| argument-hint | [file-or-pattern] |
| effort | medium |
Run the full verify suite and report the results. Do not fix anything — report only.
Primary use case: run this skill when CI is red, after a merge/rebase, or before starting a new PRD to confirm the baseline is clean. This is intentionally read-only — no fixes are applied here.
Use a subagent to run the commands so the output does not consume the main session's context. The subagent should return a structured summary.
Steps:
tsc --noEmit, mypy, pyright, or equivalent — skip if none detected)
b. Lint (e.g. eslint, ruff, clippy, or equivalent — skip if none detected)
c. Tests (e.g. vitest, pytest, cargo test, go test ./..., or equivalent)
d. Build (e.g. npm run build, cargo build, go build, or equivalent — skip if none detected)ACTION REQUIRED — do not end your response without doing this:
If running as a subagent (no direct user interaction), skip the question and return the structured summary instead.
/2_review <spec-name>?"/3_fix if these are known review issues B) Route to /debug for a specific error C) Show me the full error output first"Do not end your response without asking this question.