Skill 2 of 6 · Critique

deep-plan-review

/deep-plan-review

Fresh agents critique your finished plan against your codebase and your recorded preferences - catching problems while they're still cheap to fix.

What it does for you

A second opinion that owes the plan nothing

Authors are too close to their own plans to see the blind spots. This review isn't.

Fresh eyes, real independence

Reviewers never see the planning conversation - so they can't inherit its framing, its rationalizations, or its blind spots. They judge the plan against ground truth.

Two lenses, two sources of truth

The plan is checked against you - your taste, preferences, and recorded decisions - and against your codebase - patterns, duplicated behavior, and conflicts.

Findings you can act on

Every finding is severity-ranked and carries evidence plus a concrete recommendation. Clean lenses are called clean - no findings invented to look thorough.

Scales to the plan

One fresh agent covers most plans; for large, multi-phase work, a parallel multi-agent fan-out reviews every dimension independently, then synthesizes one clean report.

Why it pays

The cheapest catch in the whole pipeline

A misaligned plan caught here costs a document edit. The same problem caught at code review costs a rewrite - and caught in production, far more.

No hand-waving allowed: every finding must cite a file path, a plan section, or a named existing function - vague "consider improving" advisories are banned outright. Accepted fixes are folded back into the plan so nothing gets lost.

Where it fits

The gate between planning and building

Nothing gets implemented until an independent reviewer has had its say.