name: cortex-review description: Code and design review grounded in persistent memory — compares new work against accumulated knowledge and patterns

Cortex Review

Review code, designs, or proposals by comparing them against your accumulated knowledge in cortex.

Workflow

1. Ground in memory

Before reading the work under review, query cortex for the topic:

query("the domain or system being reviewed")

Read the results. They contain past decisions, patterns, and lessons learned that inform your review.

2. Review against context

As you read the work, compare it against what cortex returned:

• Does it align with established patterns and past decisions?
• Does it diverge from known approaches — intentionally or accidentally?
• Does it introduce novel patterns worth capturing?

3. Record what you find

New patterns: If the work introduces something worth remembering:

observe("The new caching layer uses write-through with 5min TTL — effective for this read-heavy workload")

Open questions: If something isn't clear:

wonder("Why did they bypass the rate limiter for internal services?")

Belief updates: If the work changes your understanding:

believe(concept_id, "Updated understanding", "Evidence from this review")

4. Output format

## Review — Grounded in Memory

### Aligned with known patterns
- [what matches cortex context]

### Divergences
- [what differs, with reasoning about whether intentional]

### New patterns to capture
- [novel approaches worth an observe() call]

### Open questions
- [things to wonder() about]

Why This Matters

Without memory grounding, every review starts from zero. You'll miss that "we tried this approach 3 weeks ago and it caused latency spikes" or "this pattern was explicitly chosen over the alternative for compliance reasons." Cortex holds that context.