cross-modal-review

Cross-Modal Review — Second Opinion Quality Gate

Send work to a different AI model for independent review. The reviewer grades against the originating skill's Contract section — checking promises, not vibes.

Contract

This skill guarantees:

• Work product is reviewed by a different model before finalizing
• Review grades against the originating skill's Contract (what was promised)
• Agreement and disagreement reported transparently
• User always makes the final decision (user sovereignty)
• Never auto-applies reviewer suggestions

Phases

Phase 1: Capture Work Product

Identify what needs review:

• The brain page, code change, analysis, or decision
• The originating skill (which skill produced this output)
• The Contract section from that skill (what was promised)

Phase 2: Load the Contract

Read the originating skill's Contract section. This is the grading rubric. Example: if the output came from gstack-review, load gstack-review/SKILL.md and extract the Contract section.

If no originating skill (ad-hoc work), construct a minimal contract:

• What was the user's request?
• What did the agent promise to deliver?
• What quality bar was implied?

Phase 3: Spawn Review Model

Use Cursor's Task tool to spawn a subagent with a different model:

Task(
  model: "fast" or a named alternative model,
  prompt: "You are an independent reviewer. Grade this work against the Contract below.
  
  CONTRACT:
  {paste the Contract section}
  
  WORK PRODUCT:
  {paste the work to review}
  
  For each promise in the Contract, answer: PASS or FAIL with specific evidence.
  Then give an overall verdict: PASS / ISSUES FOUND.
  List specific findings with evidence."
)

Phase 4: Synthesize

Present the review to Kevin:

Cross-Modal Review
==================
Reviewer: {model name}
Contract: {originating skill name}
Verdict: PASS | ISSUES FOUND

Contract compliance:
  [PASS] {promise 1} — {evidence}
  [FAIL] {promise 2} — {what's missing}

Findings:
  1. {finding with evidence}
  2. {finding with evidence}

Agreement with primary: {X}%
Recommendation: {accept / revise / redo}