← ClaudeAtlas

project-auditlisted

Perform a deep, evidence-based, read-only audit of a software project. Use when Codex needs to map architecture, build and deployment paths, server posture, security risks, unfinished work, and prioritized remediation before proposing or making changes.
bglglzd/agent-workflows · ★ 1 · AI & Automation · score 72
Install: claude install-skill bglglzd/agent-workflows
# Project Audit Audit the project before changing it. Produce conclusions that a maintainer can verify from evidence. ## Workflow 1. Inventory tracked files and relevant untracked configuration. Identify stack, entry points, package manifests, tests, docs, scripts, CI/CD, containers, infrastructure, environment templates, and git state. 2. Map the system: application and data flow, trust boundaries, auth and permissions, persistence, background jobs, external APIs, and user-facing surfaces. 3. Map operations: configuration, local setup, build, test, deploy, monitoring, backup, rollback, and any server connection instructions found in the repository. 4. Review security using the actual stack: secrets, input validation, authn/authz, dependency and supply-chain risk, file/network access, logging, errors, CI permissions, containers, and exposed configuration. 5. Review engineering health: test coverage, reliability, performance hotspots, maintainability, documentation, TODOs, migrations, release process, and unfinished product work. 6. Run only safe and relevant existing checks. Record each command and outcome; identify checks that could not run and why. ## Safety boundary - Keep the audit read-only unless the user explicitly approves a fix. - Do not connect to servers, use credentials, call production APIs, scan networks, or run destructive commands without separate explicit authorization. - Do not reveal secrets or private data. Cite a redacted location instead. - Separate