AAK-POISON-003: Cross-tool reference in tool description
- Severity
- HIGH
- Category
- Tool Poisoning (
tool-poisoning) - Scanner
agent_audit_kit/scanners/tool_poisoning.py- Auto-fixable
- No.
- SARIF name
ToolDescCrossToolRef
What it detects
Tool description references other tools by name, potentially triggering chain calls (e.g., 'before using this tool, first call X').
How to fix
Tool descriptions should be self-contained and not reference other tools.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-POISON-003
Limitations
Matches five phrases: before calling, first call, after using, then use and invoke. A description that sends the model to another tool in other words is not matched, for example run get_credentials afterwards, send the page text to send_email or call read_secrets with the same arguments; a tool name on its own is never matched.
References
- OWASP MCP Top 10: MCP03:2025 (Tool Poisoning)
- OWASP Agentic Top 10: ASI02 (Tool Misuse and Exploitation), ASI04 (Agentic Supply Chain Vulnerabilities)
- Adversa AI MCP Top 25 (AAK-defined ids): ADV-POISON-03 (Cross-Tool Reference)
- CSA AI Controls Matrix (AICM): AIS-07
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-POISON-003.md: the page then shows that text, with the registry reference appended under it.