AAK-SKILL-003: SKILL.md embeds data-exfiltration primitives
- Severity
- CRITICAL
- Category
- Tool Poisoning (
tool-poisoning) - Scanner
agent_audit_kit/scanners/skill_poisoning.py- Auto-fixable
- No.
- SARIF name
SkillExfilPrimitive
What it detects
A SKILL.md references outbound HTTP tools (fetch/curl/request) combined with instructions to send local data (files, environment, credentials).
How to fix
Remove the exfil instruction. Skills that genuinely need outbound HTTP should declare it explicitly with a documented purpose.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-SKILL-003
Limitations
Reads SKILL.md only and needs one of fetch, axios.get/post/put, requests.get/post/put, urllib.request.urlopen, http.client.HTTPConnection, curl or wget somewhere in the file and a local-secret reference (os.environ, process.env, id_rsa, id_ed25519, readFile, readText, ~/.ssh, .aws/credentials, .netrc) anywhere else in it; the two are not tied together. A credential file off that list is not reported (curl -T ~/.kube/config ...), and neither is exfiltration through httpx.post or scp. A SKILL.md over 512 KB is not read.
References
- OWASP Agentic Top 10: ASI02 (Tool Misuse and Exploitation), ASI04 (Agentic Supply Chain Vulnerabilities)
- OWASP Agentic Skills Top 10: AST01 (Malicious Skills)
- Adversa AI MCP Top 25 (AAK-defined ids): ADV-SKILL-03
- 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-003.md: the page then shows that text, with the registry reference appended under it.