hellblazer/code-review

Code Review Skill

Delegates to the code-review-expert agent.

Model Selection

Default: haiku. Escalate via model parameter on the Agent tool:

| Task Shape | Model | When | |-----------|-------|------| | Small diff, routine review | haiku (default) | <200 LOC, no security concerns | | Security-sensitive, >500 LOC, or architectural | sonnet | Auth code, crypto, API boundaries |

When This Skill Activates

• After writing or modifying significant code (10+ lines)
• When completing a feature or bug fix
• After refactoring existing code
• Before creating a pull request
• When code quality, security, or best practices review is needed

digraph review_flow {
    "Code changes ready?" [shape=diamond];
    "Run tests first" [shape=box];
    "Invoke code-review-expert" [shape=box];
    "Critical findings?" [shape=diamond];
    "Fix and re-review" [shape=box];
    "Invoke test-validator" [shape=doublecircle];

    "Code changes ready?" -> "Run tests first" [label="yes"];
    "Run tests first" -> "Invoke code-review-expert";
    "Invoke code-review-expert" -> "Critical findings?";
    "Critical findings?" -> "Fix and re-review" [label="yes"];
    "Critical findings?" -> "Invoke test-validator" [label="no"];
    "Fix and re-review" -> "Invoke code-review-expert";
}

Pre-Dispatch: Seed Link Context (optional)

If the review references an RDR or bead, seed link-context so any patterns the agent stores to T3 auto-link. See /nx:catalog for details. Skip if the review is purely ad-hoc.

Agent Invocation

Use the Agent tool to invoke code-review-expert:

## Relay: code-review-expert

**Task**: [what needs to be done]
**Bead**: [ID] or 'none'

### Input Artifacts
- Files: [relevant files]

### Deliverable
Structured code review with severity-rated findings

### Quality Criteria
- [ ] All changed files analyzed
- [ ] Security vulnerabilities flagged
- [ ] Specific remediation guidance provided

For full relay structure and optional fields, see RELAY_TEMPLATE.md.

Review Methodology

The code-review-expert agent uses hypothesis-driven review:

1. Form hypothesis about code quality patterns
2. Gather evidence from code structure, naming, patterns
3. Validate against best practices and security requirements
4. Document findings with file:line references

REQUIRED BACKGROUND: Use /nx:receiving-review when acting on review output.

Agent-Specific PRODUCE

• Session Scratch (T1): scratch tool: action="put", content="<notes>", tags="review" — working review notes during session; flagged items auto-promote to T2 at session end
• nx memory: memory_put tool: content="...", project="{project}", title="review-findings.md" — persistent review findings across sessions
• nx store (optional): store_put tool: content="...", collection="knowledge", title="pattern-code-{topic}", tags="pattern,code-review" — recurring violation patterns worth long-term storage
• Beads: creates bug beads (/beads:create "..." -t bug) for critical findings that require follow-up work

Success Criteria

• [ ] All changed files analyzed
• [ ] Security vulnerabilities flagged
• [ ] Best practices validated
• [ ] Specific remediation guidance provided
• [ ] At least one positive feedback item included
• [ ] T2 memory updated with session findings (if multi-session work)

Session Scratch (T1): Agent uses scratch tool for ephemeral working notes during the session. Flagged items auto-promote to T2 at session end.

On Completion (Mandatory)

On successful review completion, write a T1 scratch marker so the PreToolUse verification hook can confirm review happened this session:

nx scratch put "review-completed bead={bead-id} at={ISO-timestamp}" --tags "review,{bead-id}"

Replace {bead-id} with the bead ID from the relay (e.g., nexus-4yit). Replace {ISO-timestamp} with the current UTC time in ISO 8601 format (e.g., 2026-04-01T16:00:00Z).

No bead context: If invoked without a bead ID (ad-hoc review), write the marker with bead=none:

nx scratch put "review-completed bead=none at={ISO-timestamp}" --tags "review"

The --tags flag format is a comma-separated string: --tags "review,{bead-id}" (not --tags review --tags {bead-id}).