Agent Skills: Review Plan

Review Plan

• Full Review gate for Shaping, Modeling, Slicing, and Plan Prompts under docs/shaping/[feature].
• Each Subagent receives the same artifact content and reports only Critical, Important, or Minor findings.

1. Read the artifacts

The Architect provides a feature name or file path. Determine which files exist and read every available artifact before dispatching Subagents.

• shaping.md contains requirements, boundaries, shapes, and fit checks.
• affordances.md contains User Interface Affordances, Code Affordances, data stores, and wiring.
• slices.md contains Slice definitions with acceptance criteria.
• V*-plan.md contains Slice implementation Plans.

Every reviewer receives the same artifact content

Paste the full content from every available artifact into each Subagent Prompt.

2. Build the five reviewer Prompts

Use agent-prompts.md to build the five review-plan Subagent Prompts.

• Completeness uses subagent_type: "code-reviewer" and checks requirement tracing, acceptance criteria, code-claim Verification through /trace, assumptions, infrastructure prerequisites, consumer propagation, research depth, gaps, and scope proportionality.
• Critical Path uses subagent_type: "ux-tester" and checks whether every new User-facing feature has an entry point, steps, exit, states, and reachability.
• Regressions uses subagent_type: "code-reviewer" and checks existing behavior, callers through /trace, preservation, tests, shared utilities, bidirectional impact, cross-system cascades, and migration completeness.
• Architect uses subagent_type: "architect" and checks Precedents, Claude.md Rules, encapsulation, dependency direction, reuse, Modeling quality, library Verification, Decision completeness, ordering, and behavior changes.
• Frontend uses subagent_type: "frontend-engineer" and checks frontend Architecture, Claude.md Rules, sibling components, reuse, data movement, component quality, and interactive states.

3. Dispatch five Subagents in parallel

Launch all five reviewer Subagents through the Agent tool with the Prompts from step 2.

4. Aggregate the findings

Prefix each issue with the Subagent name and group by severity.

Template: # Plan Review

## Critical
[issues from all Subagents, prefixed with Subagent name]

## Important
[issues from all Subagents, prefixed with Subagent name]

## Minor
[issues from all Subagents, prefixed with Subagent name]

If all clear: "No issues found."

5. Apply the gate

Critical blocks until artifacts are revised.

Important is reported to the Architect.

Minor is reported and does not block.