cavecrew
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the injection of data from subagents back into the main agent's context. These subagents ingest untrusted external data such as source code files and diffs.
- Ingestion points: Tool-results from subagents (investigator, builder, and reviewer) are merged into the main conversation context as specified in SKILL.md.
- Boundary markers: The output contracts in SKILL.md provide structure but lack explicit delimiters to prevent the main agent from potentially following instructions embedded within the reported data.
- Capability inventory: The skill manages subagents with the ability to read from the file system and perform surgical code edits.
- Sanitization: The documentation does not specify any logic for sanitizing or filtering external content before it is returned to the main agent context.
Audit Metadata