跳到正文
cccskills
免费在 GitHub 公开

spec-executor

Execute approved, bounded work in an isolated implementation conversation and return a structured evidence receipt. Use with inherited history or an explicitly identified source; no particular planning, testing, or review skill is required.

中文简介正在准备中。当前显示原文说明。

查看安装方法

包含的文件(2)

  • SKILL.md5.4 KB
  • agents/openai.yaml264 B

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Spec Executor

Implement one approved unit of work in the isolated execution conversation. Treat its approved source as authoritative, keep implementation detail here, and return a compact receipt that can be pasted back.

Resolve the execution contract

  1. Resolve one approved source from inherited conversation history or the user's explicit reference. Later user corrections override older conflicting requirements.
  2. Read the complete source, including directly linked decisions. Do not reconstruct a second long goal.
  3. Confirm that the work is bounded enough for this implementation context and contains no unresolved product decisions. If not, stop and recommend a split or clarification without requiring a particular upstream workflow.
  4. Extract and preserve:
    • repository and starting baseline;
    • user-visible outcome and acceptance criteria;
    • agreed validation seam;
    • explicit non-goals and workspace boundaries;
    • authorization for commit, push, review request, deployment, tracker edits, data changes, and other external effects.

Do not reopen settled product questions. Pause only when the approved source is internally contradictory, conflicts materially with the repository, lacks a required external contract, or requires new authority.

Lock scope before editing

Inspect repository instructions, branch, HEAD, worktree status, recent relevant history, and the current implementation at the agreed seam. Record the actual pre-implementation HEAD as the review fixed point.

Send a short execution lock before substantive edits:

Executing: <one-sentence outcome>
Source: <approved conversation, issue, file, or spec>
In scope: <compact list>
Out of scope: <compact list>
Validation: <primary seam and required gates>
External authority: <what is and is not authorized>

Create or switch to an appropriate local branch when needed. Never discard, overwrite, or absorb unrelated dirty or untracked work. If the requested baseline has drifted, determine whether the spec is still applicable and record the current fixed point; stop if drift changes product behavior or acceptance criteria.

Implement the spec

  1. Trace each acceptance criterion to existing behavior, the smallest code seam that can satisfy it, and observable evidence.
  2. Choose an implementation and testing approach appropriate to the repository, risk, and agreed seam. Prefer behavior evidence over implementation-detail assertions.
  3. Implement the narrowest complete vertical path. Do not pull in downstream tickets, opportunistic refactors, or new product behavior.
  4. Run the smallest relevant check during development, then the repository-required broad gates in proportion to risk.
  5. Compare the finished diff with every acceptance criterion and review it against the recorded baseline. Use an available specialist review skill or tool when useful, but do not require one. Fix in-scope critical findings and resolve or explicitly report the rest.
  6. Inspect the final worktree and separate this execution's changes from anything pre-existing.

Invoking this skill authorizes in-scope local implementation and validation. It does not by itself authorize committing, pushing, opening or merging a review, changing tracker state, deploying, writing production data, calling costly real services, or messaging people. Perform those actions only when the approved source or a later user message explicitly authorizes them. Never broaden one authorization into another.

Handle blockers and context overflow

When blocked, exhaust safe read-only investigation and local validation first. Then stop with evidence and the smallest decision or authority needed from the planning thread.

If the work proves too large for one reliable implementation context, do not continue by silently dropping history. Return a partial receipt, preserve the worktree, and recommend splitting or compiling the remaining slice into a portable goal.

Return the receipt

Do not write a handoff file unless requested. End with one copy-pasteable block:

SPEC EXECUTION RECEIPT

- Conclusion: completed / partially completed / blocked
- Spec source:
- Review fixed point:
- Acceptance criteria: <each criterion with pass/fail and evidence>
- Main changes:
- Changed files:
- Branch / commit / review:
- Validation results:
- Review findings:
- Not validated or not executed:
- Risks and remaining work:
- Planning-thread decision needed:
- Final worktree state:
- External effects: <push, deploy, tracker, data, real services, messages>
- Goal / spec quality: <optional: accurate / criteria-too-vague / criteria-wrong / missing-constraint / over-scoped — plus one sentence>

Keep the receipt concise but evidence-bearing. Include exact commands, counts, identifiers, and links when they materially prove completion. Never claim a real environment, deployment, or external action that was not verified. Redact credentials, tokens, cookies, personal data, and sensitive environment identifiers before the receipt leaves the execution thread.

Goal / spec quality is optional and is not a completion condition. The execution agent may leave it blank. The planning thread or the user fills it after comparing the receipt with the actual diff.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

ask-matt

免费

Ask which skill or flow fits your situation. A router over the skills in this repo.

中文简介正在准备中。当前显示原文说明。

tt-a1i/matt-skills-with-to-goal1852026年10月10日 更新

Hand the current conversation off to a fresh background agent that picks up the work immediately.

中文简介正在准备中。当前显示原文说明。

tt-a1i/matt-skills-with-to-goal1852026年10月10日 更新

Review the changes since a fixed point (commit, branch, tag, or merge-base) along two axes: Standards (does the code follow this repo's documented coding standards?) and Spec (does the code match what the originating issue/spec asked for?). Runs both reviews in parallel sub-agents and reports them side by side. Use when the user wants to review a branch, a PR, work-in-progress changes, or asks to "review since X".

中文简介正在准备中。当前显示原文说明。

tt-a1i/matt-skills-with-to-goal1852026年10月10日 更新

Shared vocabulary for designing deep modules. Use when the user wants to design or improve a module's interface, find deepening opportunities, decide where a seam goes, make code more testable or AI-navigable, or when another skill needs the deep-module vocabulary.

中文简介正在准备中。当前显示原文说明。

tt-a1i/matt-skills-with-to-goal1852026年10月10日 更新

Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.

中文简介正在准备中。当前显示原文说明。

tt-a1i/matt-skills-with-to-goal1852026年10月10日 更新

Build and sharpen a project's domain model. Use when discussing codebase terminology, writing or editing a GLOSSARY.md, or recording or editing an ADR.

中文简介正在准备中。当前显示原文说明。

tt-a1i/matt-skills-with-to-goal1852026年10月10日 更新

查看 tt-a1i 的全部技能

报告此技能的问题