code-review

flowchart TD
    A([BEGIN]) --> B[Read task brief and acceptance criteria from .ai/tasks/]
    B --> C[Read git diff of submitted changes]
    C --> D[Check: are all modified files within agent boundary?]
    D --> E{Boundary violation?}
    E -->|Yes| F["REJECT: list boundary violations in .ai/reviews/"]
    F --> Z([END])
    E -->|No| G[Check each acceptance criterion against the diff]
    G --> H{All criteria met?}
    H -->|No| I["REJECT: list unmet criteria in .ai/reviews/"]
    I --> Z
    H -->|Yes| J[Check for regressions: build, tests, related features]
    J --> K{Regressions found?}
    K -->|Yes| L["REJECT: describe regressions in .ai/reviews/"]
    L --> Z
    K -->|No| M[Check commit message format compliance]
    M --> N{Format correct?}
    N -->|No| O["CHANGES REQUESTED: fix commit message format"]
    O --> Z
    N -->|Yes| P["APPROVE: generate review report in .ai/reviews/"]
    P --> Q[Commit approve with routing header]
    Q --> Z

Flow Steps Detail

Step: Read task brief

• •Open .ai/tasks/TASK-XXX.md
• •Extract acceptance criteria checklist
• •Note the assigned agent and file scope

Step: Read git diff

• •Run git diff pre-mortal..<agent>/<task-id>
• •Identify all modified, added, and deleted files
• •Note the scope of changes

Step: Check boundaries

• •Compare modified files against AGENTS.md ownership
• •Reference .ai/boundaries.md for the complete map
• •Flag any file outside the agent's domain

Step: Check acceptance criteria

• •Go through each criterion one by one
• •Mark as met or unmet
• •For unmet criteria, note what's missing

Step: Check for regressions

• •Verify build passes: npm run build
• •Verify tests pass: npm run test
• •Verify type check passes: npx tsc --noEmit
• •Check related features still work

Step: Check commit message format

• •Verify header: [AGENT:x] [ACTION:submit] [TASK:z]
• •Verify AGENT value matches the submitting agent
• •Verify TASK value matches the task being reviewed

Step: Generate review report

• •Write to .ai/reviews/TASK-XXX-review.md
• •Include checklist, findings, feedback, and verdict
• •Use template from .ai/templates/review.md

Verdict Outcomes

Advanced Review Patterns

Dynamic Reviewer Subagents

For specialized reviews, create a dynamic subagent at runtime:

CreateSubagent(name="security-reviewer", system_prompt="<security review prompt>")
Task(subagent_name="security-reviewer", prompt="Review TASK-X for security issues...")

Templates available in .agents/subagents/ (debugger, performance, docs, test-generator).

Agent Swarm for Batch Reviews

At sprint end, dispatch multiple reviewer subagents in parallel:

# Review 5 tasks simultaneously (K2.5 supports up to 100 sub-agents)
for task in tasks:
    Task(subagent_name="reviewer", prompt=f"Review {task}...")

See .ai/patterns/agent-swarm-parallel-review.md for the full pattern.