← ClaudeAtlas

diff-interrogationlisted

Review a human or AI-generated diff as an untrusted claim. Use when merging, committing, or accepting changes that may hide regressions, missing tests, security risk, data loss, or unexplained behavior. NOT for formatting-only diffs or already-reviewed changes with no new code.
alinafe82/cognitive-deadlift · ★ 0 · Code & Development · score 70
Install: claude install-skill alinafe82/cognitive-deadlift
# Diff Interrogation ## Purpose Force a diff to prove its behavior, test coverage, and risk profile before acceptance. ## Preserves Review judgment. ## Required Evidence - Diff or PR summary. - Test output if available. - Relevant files, issue, expected behavior, and security-sensitive areas. ## Failure Signs - The review summarizes file names instead of behavior. - Missing tests are buried. - Security, data, permission, or error-handling changes are not inspected. ## When To Use - A diff came from AI, heavy autocomplete, or an unfamiliar contributor. - The change touches behavior, data, auth, permissions, persistence, or error handling. - Tests are missing or only prove the happy path. - The developer cannot explain every meaningful line. ## When Not To Use - Formatting-only diffs. - Lockfile or generated artifact updates with separate verification. - Diffs already reviewed after the latest changes. ## Inputs Expected - Diff or PR summary. - Test output if available. - Relevant files, issue, or expected behavior. - Whether security-sensitive areas are touched. ## Output Expected ```md Behavior change: Highest-risk lines: Missing proof: Questions: Recommendation: ``` ## Process 1. Summarize the behavior change, not the file list. 2. Identify the highest-risk lines or decisions. 3. Check for missing tests, widened permissions, silent failures, data loss, and hidden coupling. 4. Ask explanation questions for unclear changes. 5. Recommend commit, revise, or rej