cavecrew
Pass
Audited by Gen Agent Trust Hub on Jun 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill manages a local wiki structure by creating and modifying files in the "llmwiki/" directory to record activity logs and drafts. This behavior is consistent with the skill's intended logging and documentation functionality.
- [PROMPT_INJECTION]: Indirect surface exists where the main agent processes outputs from subagents (investigator, builder, reviewer). While these subagents return structured data, instructions could potentially be embedded in task summaries.
- Ingestion points: Main context receives task results from "cavecrew-investigator", "cavecrew-builder", and "cavecrew-reviewer".
- Boundary markers: Not explicitly defined in the instructions for separating subagent data from main instructions.
- Capability inventory: File-write access to the "llmwiki/" directory.
- Sanitization: No specific sanitization or validation of subagent output is specified before logging to the wiki.
Audit Metadata