Validate Plan Skill

Run Phase 2 validation with Cursor and Gemini in parallel.

Overview

This skill coordinates parallel review of the implementation plan by:

•Cursor: Security and code quality focus
•Gemini: Architecture and scalability focus

Both agents must approve before proceeding to implementation.

Prerequisites

•Phase 1 (Planning) completed
•Plan available in phase_outputs (type=plan)
•Cursor and Gemini CLIs available

Execution Steps

1. Prepare Validation Storage

Ensure validation outputs will be stored in phase_outputs and logs (no local directories).

2. Read the Plan

code

Read plan from `phase_outputs` (type=plan)

3. Run Agents in Parallel

Execute BOTH Bash commands simultaneously:

Cursor Validation:

bash

cursor-agent --print --output-format json "
# Plan Validation - Security & Code Quality Review

## Plan to Review
Read: plan JSON provided in this prompt (from phase_outputs export)

## Your Focus Areas
1. **Security Analysis**
   - Identify potential vulnerabilities in proposed changes
   - Check for injection risks (SQL, XSS, command)
   - Evaluate authentication/authorization design
   - Assess data validation approach

2. **Code Quality**
   - Review proposed file structure
   - Check for maintainability concerns
   - Evaluate test coverage plan
   - Identify potential code smells

3. **Risk Assessment**
   - Highlight high-risk changes
   - Suggest additional safeguards

## Output Format
Return JSON:
{
  \"agent\": \"cursor\",
  \"phase\": \"validation\",
  \"approved\": true|false,
  \"score\": 1-10,
  \"assessment\": \"Summary of your review\",
  \"security_findings\": [
    {\"severity\": \"high|medium|low\", \"finding\": \"\", \"recommendation\": \"\"}
  ],
  \"quality_concerns\": [
    {\"area\": \"\", \"concern\": \"\", \"suggestion\": \"\"}
  ],
  \"blocking_issues\": [],
  \"strengths\": []
}
" > cursor-feedback.json

Gemini Validation:

bash

gemini --yolo "
# Plan Validation - Architecture & Scalability Review

## Plan to Review
Read: plan JSON provided in this prompt (from phase_outputs export)

## Your Focus Areas
1. **Architecture Review**
   - Evaluate design patterns chosen
   - Check component modularity
   - Assess separation of concerns
   - Review data flow design

2. **Scalability Assessment**
   - Identify potential bottlenecks
   - Evaluate performance implications
   - Check resource utilization patterns

3. **Maintainability**
   - Assess long-term maintenance burden
   - Check for technical debt introduction
   - Review extensibility

## Output Format
Return JSON in a code block:
\`\`\`json
{
  \"agent\": \"gemini\",
  \"phase\": \"validation\",
  \"approved\": true|false,
  \"score\": 1-10,
  \"assessment\": \"Summary of your review\",
  \"architecture_review\": {
    \"patterns\": [{\"pattern\": \"\", \"appropriate\": true|false, \"notes\": \"\"}],
    \"modularity_score\": 1-10,
    \"concerns\": []
  },
  \"scalability_review\": {
    \"bottlenecks\": [],
    \"recommendations\": []
  },
  \"blocking_issues\": [],
  \"strengths\": []
}
\`\`\`
" > gemini-feedback.json

4. Parse Results

Read both feedback files:

•phase_outputs entry cursor_feedback
•phase_outputs entry gemini_feedback

For Gemini, extract JSON from markdown code block if needed.

5. Evaluate Approval

Approval Criteria (Phase 2):

Criterion	Requirement
Cursor Score	>= 6.0
Gemini Score	>= 6.0
Combined Average	>= 6.0
Blocking Issues	None from either agent

6. Handle Conflicts

If agents disagree (one approves, one doesn't):

Conflict Type	Resolution
Security concern	Cursor wins (weight 0.8)
Architecture concern	Gemini wins (weight 0.7)
Both have blocking	Human escalation required

Use /resolve-conflict skill for complex disagreements.

7. Create Consolidated Feedback

Write consolidated output to phase_outputs (type=validation_consolidated):

json

{
  "validation_result": "approved|needs_changes|rejected",
  "cursor": {
    "approved": true,
    "score": 7.5,
    "blocking_issues": []
  },
  "gemini": {
    "approved": true,
    "score": 8.0,
    "blocking_issues": []
  },
  "combined_score": 7.75,
  "all_concerns": [...],
  "resolution": "Both agents approved",
  "proceed_to_implementation": true
}

8. Update State

If Approved:

json

{
  "current_phase": 3,
  "phase_status": {
    "validation": "completed",
    "implementation": "in_progress"
  },
  "validation_feedback": {
    "cursor": { ... },
    "gemini": { ... }
  }
}

If Not Approved:

json

{
  "current_phase": 1,
  "phase_status": {
    "validation": "needs_revision"
  },
  "validation_feedback": { ... },
  "iteration_count": 2
}

Return to planning with feedback to address concerns.

Retry Logic

•Max iterations: 3
•After 3 failed validations: Escalate to human
•Each retry should address previous concerns

Outputs

Output Type	Purpose
`cursor_feedback`	Raw Cursor response
`gemini_feedback`	Raw Gemini response
`validation_consolidated`	Merged evaluation

Error Handling

Agent Timeout

•Retry once with doubled timeout
•If still fails: Proceed with available agent
•Log warning in consolidated feedback

Parse Failure

•Try to extract partial JSON
•If completely unparseable: Treat as soft failure
•Request human review

Both Agents Fail

•Cannot proceed automatically
•Escalate to human with error details

Related Skills

•/plan-feature - Previous phase (planning)
•/call-cursor - Cursor agent details
•/call-gemini - Gemini agent details
•/resolve-conflict - Conflict resolution
•/implement-task - Next phase (implementation)