AAK-SKILL-002: SKILL.md uses unicode steganography in tool descriptions
- Severity
- HIGH
- Category
- Tool Poisoning (
tool-poisoning) - Scanner
agent_audit_kit/scanners/skill_poisoning.py- Auto-fixable
- No.
- SARIF name
SkillUnicodeSteg
What it detects
A SKILL.md body contains hidden characters (bidi override, zero-width, tag-unicode) that render differently than they parse. This hides malicious tool-use instructions from human review.
How to fix
Remove U+200B-U+200F, U+202A-U+202E, U+E0000-U+E007F and similar invisible / bidi characters from skill text.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-SKILL-002
Limitations
Reads SKILL.md only and matches U+200B to U+200F, U+202A to U+202E, U+2066 to U+2069, U+FEFF and the Unicode tag block (U+E0000 to U+E007F). Other invisible code points are not matched, such as the word joiner (U+2060) and the variation selectors (U+FE0F, U+E0100 to U+E01EF) that can carry smuggled text. 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: AST01 (Malicious Skills)
- Adversa AI MCP Top 25 (AAK-defined ids): ADV-SKILL-02
- 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-002.md: the page then shows that text, with the registry reference appended under it.