mcp-security-auditor
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes
mcp-audit.jsonl, which contains arbitrary input and output data from MCP servers. This data is untrusted as it originates from external interactions and could contain malicious instructions intended to influence the agent during the audit process. - Ingestion points:
mcp-audit.jsonl(referenced in SKILL.md). - Boundary markers: The skill projects data into TOON format using
scripts/json-to-toon.cjs, but does not define explicit delimiters to prevent the agent from following instructions embedded in the log data. - Capability inventory: The skill has access to
Bash,Read,Glob, andGreptools. - Sanitization: The skill utilizes
scripts/security/sanitize-mcp-input.shto redact authorization tokens and truncate large inputs, which provides some protection but may not filter against natural language prompt injection. - [COMMAND_EXECUTION]: The skill uses the
Bashtool to execute local processing scripts, specificallyscripts/json-to-toon.cjsandscripts/security/sanitize-mcp-input.sh, to transform and sanitize log data for analysis.
Audit Metadata