メニュー
Create or revise a guarded, auditable implementation spec for Codex or another agent. Use only when the user explicitly invokes $goal-ready-spec or specifically asks to make a spec goal-ready.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
Run a bounded multi-provider review and local repair workflow for a pull request, branch, or local diff, using fresh Codex exec threads for fixes and requiring separate authorization for every repository or PR publication action.
Clone and explore external codebases using the btx CLI. Use this skill when: - User wants to explore a GitHub repository (e.g., "explore anthropics/claude-code") - User asks how a library/framework works and you need to read its source - User wants to understand implementation details of an external project - User provides a repo URL or owner/repo and wants you to examine it - You need GitHub repository metadata (ownership, archived, last push) without curl
Address actionable GitHub pull request review feedback. Use when the user wants to inspect unresolved review threads, requested changes, or inline review comments on a PR, then implement selected fixes. Use the GitHub app for PR metadata and patch context and the thread-aware prc CLI for review state and inline comments.
Read-only triage and routing for GitHub repositories, pull requests, issues, patches, and review comments through the connected GitHub app and local git. Use for general GitHub orientation or summaries before choosing a mutation, review-fix, CI, or publishing workflow.
Fetch and analyze GitHub PR conversations, review submissions, and thread-aware code feedback using the prc CLI. Use for PR comment summaries, unresolved or outdated thread inspection, requested changes, and code-linked review context. Accepts PR numbers, PR URLs, or owner/repo format.
Helper for delegating coding work or read-only analysis to OpenCode with Z.ai Coding Plan GLM 5.2.
SOC 職業分類に基づく
| name | goal-ready-spec |
| description | Create or revise a guarded, auditable implementation spec for Codex or another agent. Use only when the user explicitly invokes $goal-ready-spec or specifically asks to make a spec goal-ready. |
Create a durable implementation contract under _plans/<short-slug>/ that preserves explicit requirements, ownership, and proof of completion.
Read the request, source plan, repository instructions, and targeted code, documentation, configuration, and tests that can resolve material facts.
Read references/contract.md before classifying requirements or designing the plan folder.
Ask focused questions when unresolved intent would change scope, ownership, acceptance, or verification. Do not finalize a weak contract.
Preserve the source in original-spec.md, then create or revise spec.md, progress.md, decisions.md, and audit.md from references/templates.md.
Keep spec.md a compact controlling router. Record design and interpretation choices in decisions.md when they are not fully spelled out as binding requirements. Move phase detail, context, long decision writeups, audits, and evidence into support files only when needed, and link each from the Read Map.
Trace every binding material requirement to acceptance criteria, completion criteria, or an audit row. Preserve explicit owner, state-location, lifecycle, API-shape, and no-equivalent-substitution requirements.
Define exact verification commands and evidence. Require objective and architectural evidence, not behavior-only equivalence.
Run the hygiene check and fix every error before handoff:
python3 agents/skills/goal-ready-spec/scripts/plan_file_hygiene.py _plans/<short-slug>
Return the created paths, unresolved decisions or risks, hygiene result, and the paste-ready Set Your Goal prompt from the template.
Keep the generated spec imperative and task-specific. Inherit current user, repository, and runtime policies. Record only task-specific deviations; do not embed a generic delegation or tool-usage manual.