AAK-SKILL-001: SKILL.md contains a post-install / side-effect command
- Severity
- CRITICAL
- Category
- Tool Poisoning (
tool-poisoning) - Scanner
agent_audit_kit/scanners/skill_poisoning.py- Auto-fixable
- No.
- SARIF name
SkillPostInstallCommand
What it detects
A SKILL.md frontmatter or body declares a post-install or auto-run command (bash, curl, pipe-to-sh, wget). Skills should be declarative; arbitrary installation commands are a rug-pull risk.
How to fix
Remove the post-install command. If setup is genuinely required, document it for the user rather than auto-running it.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-SKILL-001
Limitations
Reads SKILL.md only. A download piped or substituted into a shell or interpreter (curl ... | bash, bash -c "$(curl ...)") counts anywhere it names a remote host and no guardrail forbids it; npm install -g, pip install --user, bash -c, sh -c and eval( count only where a command starts. Not matched: those words inside a sentence ("install the CLI with npm install -g notes-cli"), a download saved and then run (curl ... -o install.sh && sh install.sh), npx -y <package>, and a script shipped beside the skill (scripts/install.sh). A SKILL.md over 512 KB is not read.
References
- OWASP Agentic Top 10: ASI04 (Agentic Supply Chain Vulnerabilities), ASI05 (Unexpected Code Execution (RCE))
- OWASP Agentic Skills Top 10: AST01 (Malicious Skills)
- Adversa AI MCP Top 25 (AAK-defined ids): ADV-SKILL-01
- CSA AI Controls Matrix (AICM): STA-10
All rules. This page is generated from the rule registry when the documentation is built; to change it, change the rule in agent_audit_kit/rules/builtin.py. To give the rule a hand-written write-up, add docs/rules/AAK-SKILL-001.md: the page then shows that text, with the registry reference appended under it.