mcpField
F
full-stop

full-stop

github.com/sztadii其他09 次安装今天更新
代码仓库

关于 full-stop

仅在用户或其他技能明确调用此技能时触发

安装

复制下面的命令,在终端中运行。

$npx skills add https://github.com/sztadii/agent-skills --skill full-stop

安装后在 Claude Code 中输入/full-stop确认已加载。

附带文件agents/openai.yaml

README 原文English

来自项目 README

README 暂无中文译文,以下为英文原文;中文介绍见上方。

Full Stop

Scope

Implement a change, simplify, verify and review it once, then publish it as a PR. Invocation authorizes implementation, local setup, verification, independent review, task commits/pushes and PR publication. Publication ends this workflow; do not maintain, merge or deploy the PR.

Honor narrower constraints such as “do not push”. Resolve routine choices without phase approvals. Escalate only blocking product decisions, access problems or out-of-scope actions, after finishing independent work.

All commits and pushes must use commit-push through file-pr. Do not run separate commit/push commands outside that workflow.

Dependencies

Required: $simplify, $proof, $council, $file-pr and $commit-push.

Before starting, find each required skill in the agent's catalog or installed locations, read its SKILL.md and complete its dependency check. Repository links alone are insufficient. If any skill is missing or unreadable, stop this workflow and its caller; name the affected skills without substituting or installing.

Workflow

Deliver

  1. Infer acceptance criteria, scope, actual base and destination from context and repository instructions. Reuse the matching task branch/PR; preserve unrelated changes. Do not require tickets or update issue trackers.
  2. Implement the complete change, meaningful regression coverage and required documentation. Inspect the whole diff for readability and incidental churn; keep tests proportional to risk. Avoid redesigns.
  3. Invoke $simplify once on the complete local task diff with the check scope below. Retain its changes and verification result; a no-op is valid. Stop on failed or incomplete verification.
  4. Start $council once on the simplified snapshot with intent, $simplify results and user-selected reviewers or settings. While reviewers work, invoke $proof on the same snapshot with acceptance criteria and the check scope below; give $council the $proof result before it synthesizes findings. Keep the reviewed snapshot stable until reviewers finish.
  5. Fix substantiated $council findings and in-scope $proof failures, then rerun $proof for the affected checks. Never start another $council round; record these fixes as not council-reviewed. Unfixed findings and failed or incomplete verification leave delivery incomplete; report them.
  6. Invoke $file-pr with prepared changes, matching PR, constraints, findings and useful evidence with its snapshot. Use its publication result.

Check scope

  • $simplify and $proof run checks only for changed files and code that depends on them: lint those files, run related tests (for example vitest related or jest --findRelatedTests) and typecheck as narrowly as the tooling allows. Avoid whole-suite commands such as npm test; the PR CI runs full checks.

Pass scope, permissions, revision and evidence between skills. Preserve each skill's returned state, evidence and blockers; do not bypass blockers or repeat child actions.

Result

Report delivered behavior, evidence, $council verdict, fixes made after it, PR URL, published head and draft status; report pending CI as pending. Distinguish failed, unverified and deferred requirements. Do not invent success.

评论

支持 Markdown