bug-hunting
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill documentation in 'references/mcp-agent-security.md' explicitly lists phrases like 'ignore previous instructions', 'forget your rules', and '' blocks. These are identified as patterns the agent should flag when auditing other MCP servers and do not represent an attempt to override the agent's own safety guidelines.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze external codebases and repository data, creating an attack surface for indirect prompt injection through the ingested files. 1. Ingestion points: Source code, commit history, and dependency manifests located at 'MCP_REPO_ROOT'. 2. Boundary markers: The 'references/mcp-agent-security.md' file advises that all tool output be treated as untrusted data and suggests using fencing. 3. Capability inventory: The skill leverages scanners for file reading, git mining, and complexity analysis. 4. Sanitization: The skill identifies the absence of sanitization in audited code as a flaw but does not describe internal sanitization for the repo data it processes.
Audit Metadata