菜单
Review and validate Ito changes, specs, or implementations. Use when the user wants a quality check, code review, or validation of project artifacts.
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
Implement Ito Lite markdown changes by reading proposal/spec/design/tasks artifacts and updating manual task statuses. Use when applying or executing a prompt-only Ito Lite change without the ito CLI.
Archive completed Ito Lite changes by manually merging spec deltas into current specs and moving the change to archive. Use after implementation and verification when no ito CLI is available.
Create prompt-driven Ito Lite change proposal packages in markdown. Use for drafting proposal.md, spec deltas, optional design.md, and tasks.md without the ito CLI or validation executable.
Review Ito Lite markdown proposal packages before implementation. Use to check proposal.md, spec deltas, design.md, and tasks.md for consistency, risk, testability, and manual validation without the ito CLI.
Initialize a markdown-only Ito Lite workspace with no executables. Use when setting up prompt-driven change proposals, creating .ito-lite/project.md, or preparing a repo for Ito-style specs, changes, and tasks without the ito CLI.
Coordinate the prompt-only Ito Lite workflow for proposals, specs, tasks, reviews, implementation, and archive without the ito executable. Use when the user asks for Ito-style planning, change proposals, lightweight specs, or markdown-only change management.
| name | ito-review |
| description | Review and validate Ito changes, specs, or implementations. Use when the user wants a quality check, code review, or validation of project artifacts. |
Determine the change ID and render the authoritative review instruction:
ito agent instruction review --change "<change-id>"
Review proposal/spec compliance before implementation quality. Trace all acceptance criteria to observed code, tests, documentation, or CLI behavior. Run current verification commands; never claim success from stale output, intention, or unchecked task status.
Use independent review passes for non-trivial changes. Native test-runner or specialist agents may be delegated only where the harness exposes a genuine agent surface; they are not required skill fallbacks. Fix critical and important findings, then rerun the affected evidence.
Report findings with [blocking], [suggestion], or [note], followed by Verdict: approve, Verdict: request-changes, or Verdict: needs-discussion. Completion evidence must include the exact checks run, meaningful results, unresolved risks, and acceptance-criteria coverage.