← ClaudeAtlas

security-triagelisted

Verifies, deduplicates, ranks and patches security findings for a web/SaaS product that the user owns or is authorized to test. Takes output of security-audit, pentester, saas-entitlements, server-armor, or any scanner (Semgrep, ZAP, Snyk, Trivy, npm audit, bug-bounty reports, an old backlog) and returns a short evidence-backed list: what is real, what is a false positive and why, severity derived from preconditions, owner, and optional candidate patches that are re-verified with a failing test and a fresh re-attack. Read-only by default. Triggers: "/security-triage", "triage findings", "verify scanner output", "false positives", "dedupe findings", "prioritize vulns", "fix the findings", "patch these vulns", "go through the findings", "filter out the noise", "which of these are real", "what should I fix first".
arcticloud/mythos-security-skills · ★ 0 · Web & Frontend · score 62
Install: claude install-skill arcticloud/mythos-security-skills
# Security triage: verify, rank, patch, re-verify Finding candidate bugs is now cheap: one audit pass or one scanner run returns dozens of them. The expensive part is everything after: deciding which are real, which are the same bug, how bad each one is, and fixing them without breaking the product. This skill is that second half of the loop. ``` THREAT_MODEL ─► discovery (other skills / scanners) ─► VERIFY ─► DEDUPE ─► RANK ─► PATCH ─► RE-VERIFY ▲ │ └──────────────────── update threat model + known findings ◄────────────────────────┘ ``` **Two rules carry the whole skill:** 1. **Discovery optimizes recall, verification optimizes precision.** Never let the same pass do both. A finder that also judges quietly drops real bugs; a verifier that sees the finder's reasoning tends to agree with it. 2. **Evidence or it did not happen.** Every verdict cites `file:line`, a quoted call site, a test result, or a request/response pair from a local fixture. Prose that "sounds right" is not evidence. ## Safety contract (non-negotiable) - Only for systems the user owns or has written authorization to test. Confirm scope before anything that leaves the local machine. - **Read-only by default.** Verification is done by reading code and config. No requests to production or shared environments, no load, no replay of real payments or webhooks. - Active proof (a failing test, a request