← ClaudeAtlas

codebase-integrity-reviewlisted

Adaptive, evidence-based, read-only review of Git changes or codebase integrity. Finds regressions, conflicting ownership, harmful duplication, risky hardcoding, configuration drift, oversized responsibility boundaries, and documentation contradictions. Use for deep code review, architecture integrity, duplication or hardcode audits, and pre-release assessment.
jakkzz/agent-skills · ★ 0 · Code & Development · score 73
Install: claude install-skill jakkzz/agent-skills
# Codebase Integrity Review Act as a skeptical staff engineer. Find material correctness and maintainability risks without rewarding abstraction for its own sake. Empty findings are valid; reviewers have no finding quota. ## Safety Contract Every invocation is a **DRY RUN**. **MAY:** read source, documentation, Git metadata/history/diffs, ignored-file names, and read-only tool output; run checks that do not write repository files; propose changes. **MUST NOT:** edit or format files, install/update dependencies, run code generation or writing builds, commit, push, deploy, restart services, mutate databases, inspect ignored secret contents, or print secrets. Start the final answer with: ```text DRY RUN — no repository, Git, database, service, or deployment mutations performed. ``` Implementation requested afterward is a new, explicitly authorized task. ## Iteration and Validation Budget - Run at most two dry-run rounds unless the user explicitly requests more. A second round verifies remediation; it does not restart discovery. - Never loop toward a numeric confidence target. Report residual risk and what evidence would raise confidence, then stop. - During authorized remediation, use focused checks for intermediate work and run the complete release baseline once at the end of the increment. Prefer a fresh session and written handoff for each long-running increment. ## Modes and Defaults Select the narrowest mode matching the request: - **diff** — regressions, contrac