security-auditor-pro
Skill0 starsChanged 13 days agoMITView on GitHub
Core procedure for authorized, evidence-driven security and functional audits with an 8-agent team (coordinator, code reviewer, database reviewer, pentest, visual agent, observer, analyst, safety supervisor), approval-gated remediation and an internal coverage-qualified 1-5 star rating. Use for security audits, vulnerability reviews, OWASP/CWE code review, auth/API/DB/TLS configuration review, functional or visual QA of the user's own authorized application, or when security-audit, functional-audit or full-audit is invoked.
In this repository4 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- full-auditSkill
- functional-auditSkill
- security-auditSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 6 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 35
Bounded live validation (optional, approval required): Only with explicit approval and within scope: pentest and/or visual-agent, with …
- Distinctive: line 36
Before each live test, show the user this proposal and wait for an explicit answer:
- Distinctive: line 54
Only assets the user owns or is authorized to test.
- Distinctive: line 55
No brute force, credential stuffing, password spraying, DoS, evasion, persistence, exfiltration, destructive SQL, uncontrolled fuzzing or …
- Distinctive: line 56
No changes to code, data or infrastructure without explicit approval for that change.
- Distinctive: line 57
Stop immediately on 5xx spikes, latency >2x baseline, resource pressure, lock contention or any unexpected write.
- Distinctive: line 58
Never claim a test ran without evidence.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Claude Security Auditor Pro: core procedure
This folder is the kit. Paths below are relative to this skill's directory:
| File | Purpose |
references/POLICY.md | Mandatory operating policy. Read before every audit. |
references/OPERATIONS.md | Step-by-step operating procedure, limits and stop conditions |
references/RATING.md | Internal 0-100 score and star rules |
references/REPORT_TEMPLATE.md | Report structure |
references/scope.example.json | Scope file template |
scripts/audit_manager.py | Local metadata helper (no network, no target access) |
AM below means python <this skill dir>/scripts/audit_manager.py (use python3 on macOS/Linux).
Modes
| Mode | Entry | Covers |
| SECURITY | security-audit | source (SAST), secrets, dependencies, auth/session, authorization, APIs, DB grants/queries, TLS/config, bounded live validation |
| FUNCTIONAL | functional-audit | user journeys, UI vs server authorization, error states, timeouts, accessibility, visual regressions |
| FULL | full-audit | both, correlated, plus rating |
Procedure
- Authorization & scope (blocking). Read
references/POLICY.md. Confirm with the user: the asset is theirs or they are authorized, environment (prefer staging), authorized hosts, test accounts, data sensitivity, allowed modes, rate limits. Record them by copyingreferences/scope.example.jsoninto the project, filling it in, and runningAM init --scope <file>. No live test happens without a recorded scope. - Capability check. Detect which tools really exist in this session (browser/Playwright MCP, Semgrep, Gitleaks, Trivy, ZAP, testssl.sh, DB access, monitoring). Never simulate a tool; list missing ones as coverage gaps.
- Baseline.
git status, app health, sanitized config, current error/latency levels. - Parallel read-only review. Delegate independent read-only tasks to the agents (if available):
code-reviewer,database-reviewer. Give each agent the scope and the policy rules in its task. - Bounded live validation (optional, approval required). Only with explicit approval and within scope:
pentestand/orvisual-agent, withobserverwatching andsafety-supervisorholding veto power. Defaults: ≤1 request/second, ≤2 concurrent. Before each live test, show the user this proposal and wait for an explicit answer:LIVE TEST PROPOSAL (requires approval) Agent: <pentest | visual-agent> Target: <URL/host> (in scope ✔ / ✖) Goal: <what it confirms, linked to file:line or finding ID> Method: <exact minimal action, test accounts used> Limits: <req/s · concurrency · timeout> Safety: observer baseline <captured?> · safety-supervisor: <APPROVED | VETO + reason> Approve? (yes / no / modify) - Analysis.
analystdeduplicates, separates evidence from hypothesis, assigns severity + confidence. Record each finding:AM add --title ... --severity ... --component ... --location ... --evidence "<redacted>" --confidence ..., thenAM update <ID> --status confirmedonce verified. - Present & decide. For every confirmed issue show evidence, impact, confidence and options: investigate · bounded validation · propose patch for approval · defer · false positive (with justification). Never patch before explicit per-change approval.
- Remediate & re-test (approved only). Apply the approved patch, run tests plus security and visual regression, compare with baseline, keep rollback steps,
AM update <ID> --status fixed. - Rate & report. Score each dimension 0-100 (or
naif not tested) perreferences/RATING.md:AM rate --code .. --identity .. --db .. --infra .. --api .. --validation ... ThenAM report --md -o security-audit-report.md.
Hard rules
- Only assets the user owns or is authorized to test. If in doubt, stop and ask.
- No brute force, credential stuffing, password spraying, DoS, evasion, persistence, exfiltration, destructive SQL, uncontrolled fuzzing or CAPTCHA bypass.
- No changes to code, data or infrastructure without explicit approval for that change.
- Stop immediately on 5xx spikes, latency >2x baseline, resource pressure, lock contention or any unexpected write.
- Never claim a test ran without evidence. Never infer "secure" from an absence of findings.
- Redact secrets and personal data in all evidence and screenshots. Keep
.security-audit/out of Git. - The star rating is an internal prioritization aid, never a certification or a guarantee.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

