CCA-F: Claude Certified Architect — Foundations

By Anthropic · Question Mode

QuestionQ118

Context Management & Reliability

You are integrating Claude Code into your Continuous Integration/Continuous Deployment (CI/CD) pipeline. The system performs automated code reviews, generates test cases, and provides feedback on pull requests. You need to design prompts that deliver actionable feedback while minimizing false positives.

Your pipeline reviews every PR with a single API call using a static prompt that contains the diff and the full text of every changed file; unchanged files are excluded. Reviews are posted asynchronously and do not block PR creation. Developers report that reviews repeatedly miss bugs involving cross-file interactions—for example, when a PR renames a function's parameters, the review does not flag callers in unchanged files that still use the old argument order. Evaluation shows that cross-file bugs make up 35% of production incidents from reviewed PRs.

What is the most effective change to your review design?

Explanation

A bounded agentic review can use repository read and search tools to trace changed symbols into unchanged caller files and validate cross-file behavior before reporting a finding. This supplies the missing context behind the observed production incidents, while a turn limit constrains the investigation. Claude Code supports non-interactive automation in CI pipelines and configurable tool permissions, including read and command tools, making this design suitable for automated review workflows.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!