CodeQL Security Audit Skill

Three independent modes — identify which one the user needs and run the corresponding script.

User Intent	Mode	Script
Scan a repo / create a DB / generate SARIF	[SCAN]	scripts/scan.sh
Read SARIF / triage vulns / generate report	[AUDIT]	scripts/audit.py
Optimize or debug a .ql query file	[TUNE]	scripts/tune.py

---

[SCAN]

bash scripts/scan.sh <repo_path> [language] [output.sarif]
# language: java | javascript | python | cpp | auto (default)

The script handles: language detection → build command selection → CodeQL DB creation → security suite scan → SARIF output.

For writing custom queries, refer to the relevant language reference: references/lang-java.md / lang-javascript.md / lang-python.md / lang-cpp.md

---

[AUDIT]

python3 scripts/audit.py <results.sarif> --output exp.md

The script handles: SARIF parsing → attack surface inventory → vuln family grouping → source→sink evidence chain extraction → exp.md output.

Claude's responsibility (what the script cannot do):

• Manually assess [SUSPICIOUS] entries with no data flow — determine if they are real vulnerabilities
• Write POC requests based on business context
• Provide concrete remediation code

---

[TUNE]

python3 scripts/tune.py <query.ql>

The script outputs a tuning checklist covering seven checks: coverage, false positives, performance, and metadata completeness.

Claude's responsibility (what the script cannot do):

• Rewrite source / sink / sanitizer logic based on checklist findings
• Debug queries with no results or unexpected output — refer to references/debugging.md