llm-wiki
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes untrusted source code and configuration files to construct a wiki. Malicious instructions embedded in these files (e.g., in README files or code comments) could potentially influence the agent's summarization or actions.\n
- Ingestion points: Phase 1 (Reconnaissance) and Phase 3 (Core Articles) read various project files, including READMEs, build configurations, and source code.\n
- Boundary markers: The instructions include guidelines to be factual, use code snippets, and provide source citations, which act as logical boundaries.\n
- Capability inventory: The skill performs file reading, directory mapping, change tracking, and file writing within the .llm-wiki/ directory.\n
- Sanitization: No explicit sanitization or escaping of external code content is mentioned before it is interpolated into the wiki articles.\n- [COMMAND_EXECUTION]: The skill instructs the agent to use standard system commands like 'find' or Glob patterns and 'git log' to perform its tasks.\n
- Evidence: The Reconnaissance and Update phases in SKILL.md explicitly mention using these commands to map the directory tree and detect changes.
Audit Metadata