一键导入
execute
Use when you have an approved spec from the Discover phase. Orchestrates implementation — parallel or serial — with autonomy and guardrails.
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。
菜单
Use when you have an approved spec from the Discover phase. Orchestrates implementation — parallel or serial — with autonomy and guardrails.
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
Use before any code change — new feature, bug fix, refactor, or tech debt. Explores the problem, builds understanding through concrete artifacts, and produces a spec that gates the transition to execution.
Use at the start of every conversation and when deciding how to approach a task. Routes to the appropriate phase of the Discover → Execute → Verify flow.
Use after execution is complete, before declaring work ready to ship. Unified verification through multiple lenses: correctness, code quality, review, and architectural feel.
Use when creating, switching, syncing, or cleaning up git branches. Enforces naming conventions and the branching flow.
Use at the start of a new conversation or when working in an unfamiliar codebase. Generates a ranked map of the repository structure to orient before diving into code.
Use when creating a git commit. Guides intelligent staging, conventional commit message authoring, and pre-commit validation.
| name | execute |
| description | Use when you have an approved spec from the Discover phase. Orchestrates implementation — parallel or serial — with autonomy and guardrails. |
Build what the spec describes. Claude executes autonomously with guardrails — escalating on surprises, deviations, or ambiguity rather than guessing.
Input: An approved spec from the Discover phase (a SPRINT-N-SPEC.md or similar at the project root).
If you were dispatched as a subagent or teammate for a specific task, stop here. Do not activate the full workflow. You have a focused job — do it and report back.
| Situation | Strategy |
|---|---|
| 3+ independent file changes | Parallel agents (team or subagents) |
| Tightly coupled changes | Serial execution |
| < 3 tasks total | Just do it — no orchestration overhead |
Use TeamCreate or the Agent tool to dispatch parallel work:
Work through tasks in dependency order:
Autonomy with guardrails. Execute the spec without asking permission for every step, but stop and ask when:
Don't guess through ambiguity. Ask.
After each logical group of changes:
code-simplifier agent on modified files to refine code for clarity, consistency, and maintainabilityWhen something unexpected comes up during execution:
Only loop back to full Discover if the surprise fundamentally changes the problem (rare).
When dispatching work to agents:
One task per agent. Don't give an agent a grab bag of loosely related work.
Execute is done when all items in the spec's file change list are implemented. The code compiles, tests pass for the changed files, and the implementation matches the spec (including any amendments).
Next: Use the verify skill for final verification before declaring the work ready to ship.