review

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

Step 1 — Gather Context

Read these to understand what the agent did:

1. kapi/sprints/*/tasks.md — find the task description
2. git log --oneline -10 — see recent commits related to this task
3. kapi/snapshot.yaml — check which agent was assigned this task
4. kapi/decisions.yaml — read existing decisions to get the next REV-NNN id

Step 2 — Ask the Human

Ask these questions one at a time. Wait for each answer before proceeding.

1. Which agent did this work? (e.g. dev, captain, dev2, tester)
2. Result: approve, reject, or edit?
   • approve — work is correct, no changes needed
   • reject — work is wrong, needs to be redone
   • edit — work was mostly right but you made corrections
3. Category? (e.g. architecture, code-quality, testing, ui, docs, refactoring)
4. Any notes? (optional — what was good, what was wrong, what you changed)

Step 3 — Write the Decision Record

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.

Step 4 — Confirm

Print:

Review recorded: [result] for [task] by [agent]
Category: [category]
File: kapi/decisions.yaml
View at: /decisions

Rules

• Never skip the human prompts — this is HITL, the human's judgment is the point
• Never auto-approve — even if the work looks fine, ask the human
• Append only — never modify or delete existing decisions
• If kapi/decisions.yaml doesn't exist, create it with the schema header from kapi/sprints/v1/tasks.md T05