AAK-SKILL-005: SKILL.md frontmatter contains prompt-injection triggers
- Severity
- HIGH
- Category
- Tool Poisoning (
tool-poisoning) - Scanner
agent_audit_kit/scanners/skill_poisoning.py- Auto-fixable
- No.
- SARIF name
SkillFrontmatterInjection
What it detects
A SKILL.md frontmatter or YAML header embeds phrases targeted at the loading model ('ignore previous', 'you are now', 'system:'). These are prompt-injection probes rather than legitimate skill metadata.
How to fix
Remove the injection trigger. Skill frontmatter should only contain name, description, tags, and similar metadata.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-SKILL-005
Limitations
Reads the frontmatter of SKILL.md (up to the closing ---, or the first 500 characters) for ten fixed phrases such as ignore previous, you are now, disregard prior and system:. The same instruction in other words (ignore all previous instructions, disregard all prior rules) is not matched, and the body of the skill is not read by this rule. A SKILL.md over 512 KB is not read.
References
- OWASP Agentic Top 10: ASI01 (Agent Goal Hijack), ASI04 (Agentic Supply Chain Vulnerabilities)
- OWASP Agentic Skills Top 10: AST04 (Insecure Metadata)
- Adversa AI MCP Top 25 (AAK-defined ids): ADV-SKILL-05
- 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-005.md: the page then shows that text, with the registry reference appended under it.