skill-security-auditor · borghei — Claude Code skill for auditing agent skills
skill-security-auditor — a skill for Claude Code, Codex, Cursor, Copilot, Gemini CLI and OpenCode that scans agent skills before install. It detects prompt injection, dangerous code, credential harvesting and supply chain risks.
LicenseRef-MIT-Commons-Clause 146 installs ★ 1,183 updated 2026-11-08
npx skills add borghei/Claude-Skills --skill skill-security-auditor -a claude-codeWhat skill-security-auditor does
Python files are checked for risky patterns: os.system, subprocess with shell=True, eval/exec, obfuscation via base64 or chr(), network exfiltration, reads from ~/.ssh or ~/.aws. Markdown, text and YAML get scanned for prompt injection: system-prompt overrides, zero-width characters, HTML comments with hidden instructions, and prompts that tell the assistant to conceal actions.
A separate checker looks at dependencies. Typosquatted package names (reqeusts), unpinned versions in requirements*.txt and package.json, inline pip/npm install calls. Pinned versions are not checked against CVE feeds; run pip-audit, npm audit or OSV-Scanner for that.
Each of the three scripts returns PASS / WARN / FAIL. PASS means "nothing obvious found", not "safe": the scanners match regexes line by line, and string building or helper indirection defeats them. After a PASS on an untrusted skill, read the code yourself.
What it helps with
- scan a skill before installing it from an untrusted source
- catch hidden instructions in markdown and YAML
- detect reads of SSH keys or AWS credentials
- spot typosquatted packages and unpinned dependency versions
- gate skill installs in CI/CD
Example prompts for the agent
- Scan
assets/sample_malicious_skill/with all three checkers - Run
python scripts/code_scanner.py skills/ --strict --jsonand show HIGH findings only - Check
requirements.txtfor unpinned versions and typosquatted package names
How to install skill-security-auditor
1Skills CLI (Vercel)
npx skills add borghei/Claude-Skills --skill skill-security-auditor -a claude-code2Manually: put the files in place
~/.claude/skills/skill-security-auditor/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/skill-security-auditor/
1Skills CLI (Vercel)
npx skills add borghei/Claude-Skills --skill skill-security-auditor -a codex2Manually: put the files in place
.agents/skills/skill-security-auditor/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add borghei/Claude-Skills --skill skill-security-auditor -a cursor2Manually: put the files in place
.cursor/skills/skill-security-auditor/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install borghei/Claude-Skills engineering/skill-security-auditor/SKILL.md2Manually: put the files in place
.github/skills/skill-security-auditor/SKILL.md1Skills CLI (Vercel)
npx skills add borghei/Claude-Skills --skill skill-security-auditor -a gemini-cli2Platform CLI
gemini skills install https://github.com/borghei/Claude-Skills --path engineering/skill-security-auditorManually: put the files in place
.devin/skills/skill-security-auditor/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add borghei/Claude-Skills --skill skill-security-auditor -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about skill-security-auditor
How do I install skill-security-auditor in Claude Code?
Run npx skills add borghei/Claude-Skills --skill skill-security-auditor -a claude-code. The skill file lands at ~/.claude/skills/skill-security-auditor/SKILL.md.
What does skill-security-auditor do?
It runs static analysis on agent skills before install: dangerous constructs in Python, prompt injection in markdown and YAML, and dependency issues. The output is a PASS / WARN / FAIL verdict with findings grouped by severity and remediation notes.
Does PASS mean the skill is safe?
No. The scanners match regular expressions line by line and only catch common, unobfuscated attack forms. String concatenation, encoding and indirection through helpers evade them. For an untrusted skill, read the scripts yourself even after a PASS.
Similar skills
All skills for Security →skill-security-auditor by other authors comes first.