菜单
Rate an agent's work. Records approve/reject/edit decision in kapi/decisions.yaml for HITL competence tracking.
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
| name | review |
| description | Rate an agent's work. Records approve/reject/edit decision in kapi/decisions.yaml for HITL competence tracking. |
| argument-hint | [task e.g. T03] |
| allowed-tools | Read, Grep, Glob, Write, Bash(git log:*), Bash(git diff:*) |
You are recording a human review of agent work. This builds the HITL decision dataset that the competence engine (v2) will consume.
Task: $ARGUMENTS
Read these to understand what the agent did:
kapi/sprints/*/tasks.md — find the task descriptiongit log --oneline -10 — see recent commits related to this taskkapi/snapshot.yaml — check which agent was assigned this taskkapi/decisions.yaml — read existing decisions to get the next REV-NNN idAsk these questions one at a time. Wait for each answer before proceeding.
Read kapi/decisions.yaml, then append a new review entry to the decisions: array.
The new entry format:
- id: REV-NNN
type: review
title: "Review: [task title from tasks.md]"
result: [approve|reject|edit]
date: "YYYY-MM-DD"
agent: [agent name]
task: [task id e.g. T03]
sprint: [sprint id e.g. v1]
category: [category from human]
notes: >
[notes from human, or empty string if none]
Compute the next REV-NNN by counting existing type: review entries + 1.
Write the updated YAML back to kapi/decisions.yaml. Use the yaml package to parse and stringify — do NOT manually construct YAML strings.
Print:
Review recorded: [result] for [task] by [agent]
Category: [category]
File: kapi/decisions.yaml
View at: /decisions
kapi/decisions.yaml doesn't exist, create it with the schema header from kapi/sprints/v1/tasks.md T05Plan a sprint interactively — reads backlog and board, brainstorms with you, writes prd.md + tasks.md.
End-of-session debrief. Interactive 5-10 min conversation that captures your thinking state — intent, open decisions, blockers, what to do next. Run before stepping away.
Pick the next unchecked task from the sprint and implement it with TDD. Reads board, posts available, implements, commits.
Quick write to the blackboard. Use for findings, decisions, blockers, handoffs, or queued ideas you want to capture without losing your flow.
Pre-sprint health check — git status, build, arch drift, infra, UX audit. Run before /prd to get a go/no-go verdict.
Start-of-session briefing. Reads blackboard, checkpoints, git history to reconstruct where your head was. Run when returning after hours or days away.