How to Invoke Reviewer Agents

Step 1: Identify Reviewer

Match your artifact type to the reviewer:

Step 2: Invoke via Task Tool

<Task>
  subagent_type: general-purpose
  description: Invoke {reviewer-name}
  prompt: |
    You are the {reviewer-name} agent.

    Read the complete agent specification from:
    ~/.claude/agents/nw/nw-{reviewer-name}.md

    Review the artifact at:
    {path-to-artifact}

    Conduct peer review following the specification.
    Provide structured YAML feedback:
    - strengths
    - issues_identified (with severity)
    - recommendations
    - approval_status
</Task>

Step 3: Interpret Feedback

The reviewer returns YAML:

approval_status: "rejected_pending_revisions"
issues_identified:
  confirmation_bias:
    - issue: "Technology assumption without rationale"
      severity: "critical"
      recommendation: "Re-elicit constraints"

Actions by status:

• approved -> Proceed to handoff
• rejected_pending_revisions -> Address issues, resubmit
• conditionally_approved -> Proceed with documented caveats

After DISCUSS wave

1. Create requirements (product-owner)
2. Review requirements (product-owner-reviewer)  <- Add this step
3. Proceed to DESIGN

After DISTILL wave

1. Create acceptance tests (acceptance-designer)
2. Review tests (acceptance-designer-reviewer)  <- Add this step
3. Proceed to DELIVER

Step 1: Address Critical/High Issues

Focus on critical and high severity issues first:

issues_identified:
  confirmation_bias:
    - issue: "AWS assumption without requirement"
      severity: "critical"  # <- Fix this first

Step 2: Revise Artifact

Update your artifact to address the issues.

Step 3: Resubmit for Review

The artifact has been revised. Please re-review:
{path-to-revised-artifact}

Issues addressed:
- Removed AWS assumption, added deployment constraints section
- Quantified performance requirements (2s p95)

Step 4: Await Approval

The reviewer validates revisions and returns updated status.

Iteration limit: Maximum 2 iterations. If not approved after 2 iterations, escalate to human facilitator.

Example: Requirements Review

I need peer review on my requirements document.

<Task>
  subagent_type: general-purpose
  description: Requirements peer review
  prompt: |
    You are the product-owner-reviewer persona.

    Read: ~/.claude/agents/nw/nw-product-owner-reviewer.md

    Review: docs/feature/{feature-id}/discuss/checkout-requirements.md

    Focus on:
    1. Confirmation bias (technology assumptions)
    2. Completeness (missing scenarios/stakeholders)
    3. Clarity (vague requirements)
    4. Testability (measurable acceptance criteria)

    Provide YAML feedback with approval_status.
</Task>

Example: Acceptance Test Review

Please review my acceptance tests for happy path bias.

<Task>
  subagent_type: general-purpose
  description: Acceptance test review
  prompt: |
    You are the acceptance-designer-reviewer persona.

    Read: ~/.claude/agents/nw/nw-acceptance-designer-reviewer.md

    Review: tests/acceptance/features/checkout.feature

    Focus on:
    1. Happy path bias (error scenario coverage)
    2. GWT format compliance
    3. Business language purity
    4. Coverage completeness

    Provide YAML feedback with approval_status.
</Task>

Example: Architecture Review

Review my architecture design for feasibility.

<Task>
  subagent_type: general-purpose
  description: Architecture peer review
  prompt: |
    You are the solution-architect-reviewer persona.

    Read: ~/.claude/agents/nw/nw-solution-architect-reviewer.md

    Review: docs/feature/{feature-id}/design/architecture-design.md

    Focus on:
    1. Architectural bias (premature technology choices)
    2. ADR quality (complete decision records)
    3. Feasibility (implementation viability)
    4. Testability (component boundaries)

    Provide YAML feedback with approval_status.
</Task>