AAK-SKILL-004: SKILL.md description hijacks a trusted skill name
- Severity
- HIGH
- Category
- Tool Poisoning (
tool-poisoning) - Scanner
agent_audit_kit/scanners/skill_poisoning.py- Auto-fixable
- No.
- SARIF name
SkillNameHijack
What it detects
A skill's frontmatter name or description mimics a well-known skill ('pdf', 'docx', 'frontend-design') but the body declares unrelated / hostile instructions. Description hijacking is the ToxicSkills 2026 signature pattern.
How to fix
Rename the skill to match its actual behavior. Cross-reference the name against the first-party skill directory.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-SKILL-004
Limitations
Compares only the frontmatter name of a SKILL.md with sixteen first-party skill names, and only a name within two characters of one in length (pdfs). A name that adds a qualifier (pdf-pro, official-pdf) is not matched, nor is a skill named exactly pdf that shadows the first-party one, and a description that says to use the skill instead of pdf is never compared. A SKILL.md over 512 KB is not read.
References
- OWASP Agentic Top 10: ASI04 (Agentic Supply Chain Vulnerabilities)
- OWASP Agentic Skills Top 10: AST01 (Malicious Skills), AST04 (Insecure Metadata)
- Adversa AI MCP Top 25 (AAK-defined ids): ADV-SKILL-04
- 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-004.md: the page then shows that text, with the registry reference appended under it.