critical-plan-review — Claude Code prompt: what it does, how to use
critical-plan-review is a Claude Code prompt that red-teams an implementation plan before writing any code. The review spots edge cases, race conditions, and architectural blind spots.
no license ★ 336 updated 2026-11-14
.claude/commands/critical-plan-review.mdWhat critical-plan-review does
The prompt instructs the model to act as an adversarial senior engineer evaluating an implementation proposal. It verifies requirement coverage, error recovery paths, safe retries, and idempotency, while pruning speculative additions using YAGNI constraints.
The output delivers a formal approval decision, categorizes defects by severity, produces an amended roadmap, and lists unverified assumptions. Running this prompt in an isolated conversation or a separate subagent prevents self-confirmation bias.
What it helps with
- stress-test an implementation draft for hidden failure modes and race conditions
- strip out speculative abstractions using YAGNI principles
- verify rollback steps and idempotency before writing code
- receive a formal plan verdict with prioritized defects
Example prompts for the agent
- Review this caching architecture draft with critical-plan-review before writing code
- Run a critical-plan-review on this checkout service implementation plan
How to install critical-plan-review
Manually: put the files in place
.claude/commands/critical-plan-review.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about critical-plan-review
How to install critical-plan-review in Claude Code?
Save the prompt text to .claude/commands/critical-plan-review.md in your project root. The command will become available in your Claude Code session.
What does critical-plan-review do?
It acts as a principal engineer reviewing technical plans for edge cases, performance bottlenecks, and security gaps. The output provides a verdict, categorized findings, and a revised plan.
Why run critical-plan-review in a fresh context?
Executing the prompt in a new context or isolated subagent helps reduce self-confirmation bias. It prevents the model from blindly accepting its own drafting assumptions.