How Do PR Reviewers Use Comprehend-First for Better Reviews?
For Engineers doing code reviews with incomplete context · Based on Priscila Andre's Comprehend-First AI Coding Skill
// TL;DR
Priscila Andre's Comprehend-First AI Coding Skill transforms PR reviews from rubber-stamp approvals into genuine, context-rich evaluations. When asked to review code in a part of the codebase you don't know well, invoke the Catch Me Up prompt with Feature and Trace exploration modes to understand how the modified code works end-to-end. Build your mental model before reading the diff. This lets you catch architectural misalignments, convention violations, and subtle bugs that surface-level reviews miss — without interrupting the PR author for a walkthrough.
Why are most code reviews actually rubber stamps?
Most code reviews fail at their core purpose. The reviewer lacks sufficient context about the subsystem being modified, so they check for surface-level issues — formatting, naming, obvious bugs — and approve. The deeper questions go unasked: Does this change align with the system's architecture? Does it follow the conventions established in this subsystem? Is the data path being modified correctly end-to-end?
Priscila Andre's Comprehend-First framework solves this by making contextual preparation a standard part of the review process, not an optional extra.
How do I prepare for a PR review using Catch Me Up?
Before opening the diff, run a Catch Me Up session:
1. Declare your role: "I am a PR reviewer with partial context on this subsystem. I need to review a change to the payment processing pipeline."
2. Select modes: Use Feature to understand how the feature works end-to-end. Use Trace to follow the specific data path the PR modifies. Add Convention if you're unsure about the coding standards in this area.
3. Ask your specific question: "Walk me through how payment events flow from ingestion to storage, and identify which components this PR is modifying in that path."
4. Request a visual: Ask for a flow diagram showing the data path, with the PR's changes highlighted.
Now open the diff. You have a mental model to compare against. You can see whether the changes make architectural sense, follow conventions, and correctly handle the data flow.
What can I catch with Comprehend-First that I'd miss without it?
With a verified mental model, you catch:
- Architectural misalignments: The PR introduces a pattern that contradicts the subsystem's existing architecture.
- Convention violations: The PR uses naming or structural patterns inconsistent with the surrounding code.
- Incomplete traces: The PR modifies one part of a data path but misses a downstream component that also needs updating.
- Incorrect assumptions: The PR author assumed a function behaves one way, but your Trace-mode analysis reveals it behaves differently.
These are the bugs and design issues that slip through surface-level reviews and cause regressions weeks later.
How do I give better review feedback using this framework?
Because you've built a mental model, your review comments become substantive. Instead of "Looks good to me" or vague style suggestions, you can write:
- "Based on the data flow I traced, this change to the ingestion handler also needs a corresponding update in the storage adapter — currently the envelope format will be inconsistent between the two."
- "The convention in this subsystem is to use interceptors rather than mocks for test payloads (confirmed via Catch Me Up Architecture + Testing mode). This PR introduces a mock — is that intentional?"
This level of feedback elevates the entire team's code quality and makes your reviews genuinely valuable.
Next step: Pick a PR on your review queue right now that touches code you don't know well. Before opening the diff, run a Catch Me Up session with Feature + Trace modes. Compare the quality of your review to your usual approach.
// FREQUENTLY ASKED QUESTIONS
How long does it take to prepare for a PR review using Comprehend-First?
A focused Catch Me Up session for PR review preparation typically takes 10 to 20 minutes — less time than scheduling and waiting for a walkthrough from the PR author. The investment pays off in review quality: you catch issues that surface-level reviews miss and provide feedback that prevents regressions.
Should I tell the PR author I used AI to understand their code?
Yes, transparency is valuable. Sharing your Catch Me Up process normalizes comprehension-first workflows across the team. It also invites the PR author to correct any misunderstandings in your mental model, which further improves review quality. There is no stigma in using AI for comprehension — it's the most productive use of the tool.
What if the AI's understanding of the code contradicts what the PR is doing?
This is the most valuable signal in a Comprehend-First review. The discrepancy may indicate a bug in the PR, an outdated assumption in the AI's model, or an intentional architectural change that needs documentation. Surface it as a review comment and ask the PR author to clarify. These are exactly the issues rubber-stamp reviews miss.