caveman-help
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documentation describes tools that ingest untrusted data from external files and code reviews, while simultaneously instructing the agent to summarize these activities into a persistent local wiki.
- Ingestion points: External markdown files processed by 'caveman-compress' and pull request content analyzed by 'caveman-review'.
- Boundary markers: The instructions do not define specific delimiters or 'ignore embedded instructions' warnings to isolate untrusted input from the agent's logic.
- Capability inventory: The agent is given explicit instructions to create, write, and append to files within the 'llmwiki/' directory.
- Sanitization: No sanitization or validation of the summarized content is mentioned before it is written to the wiki index and logs, creating a surface where malicious instructions in processed files could influence the generated summary or downstream wiki updates.
Audit Metadata