← ClaudeAtlas

unslop-auditlisted

Audit a codebase for the production failures typical of AI-generated software - hardcoded secrets, disabled row level security, IDOR and missing authorization, unvalidated inputs, missing error handling and rate limits, N+1 queries and unindexed columns, hallucinated dependencies, duplicated patch-on-patch code, and missing tests or version control. Use when the user asks to audit, review, harden, or production-check a codebase, mentions vibe coding cleanup, asks "is this safe to ship", or is preparing an AI-built app for real users. Read-only - never edits code.
RuslanAMandell/UnslopMyCode · ★ 1 · Code & Development · score 72
Install: claude install-skill RuslanAMandell/UnslopMyCode
# unslop-audit ## Overview AI code generation optimizes for a working demo, not for software that survives a thousand users. This skill finds the resulting gaps and reports them with evidence: file, line, the offending snippet, and the concrete way each one fails in production. **Never edit code in this skill.** Remediation belongs to `unslop-fix`. ## Procedure ### 1. Scan The scanner detects the stack itself. Do not ask the user what they built with. ```bash python3 skills/unslop-audit/scripts/scan.py . --out .unslop/findings.json --json ``` It always exits 0 and always writes a coverage block. If `python3` is missing, say so plainly and run the reduced pass in [references/no-python-fallback.md](references/no-python-fallback.md), then record the reduced coverage in the report. Add `.unslop/` to `.gitignore` if it is not already there, and tell the user you did it. The report is a list of live vulnerabilities in their app; committing it to a public repository would publish the exploit guide. ### 2. Verify before reporting The scanner emits most findings as `SUSPECTED`. Promote one to `CONFIRMED` only after opening the file and confirming the defect is real **in context**. Discard what does not survive that read. A false positive costs more trust than a missed P3 costs coverage. When you are not sure, leave it `SUSPECTED` and say why. For every finding you keep, write the failure scenario in the user's own terms: what an attacker or a traffic spike actually does.