caveman-evidence-review
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data such as cloud reports and traces, which could contain malicious instructions if the tracked applications are compromised.
- Ingestion points: The skill fetches data through
caveman_report,caveman_trace_search, andcaveman_trace_get(SKILL.md). - Boundary markers: The skill includes a "Hard rule" (Rule 2) to avoid fetching completion or artifact payloads by default, which limits exposure, but does not provide explicit delimiters for the metadata it does process.
- Capability inventory: The skill has the ability to execute shell commands via the
cavemanCLI tool fallbacks (SKILL.md). - Sanitization: There is no mention of escaping, validation, or filtering of the external content retrieved from the cloud service.
- [COMMAND_EXECUTION]: The skill relies on the execution of a specific command-line interface tool to perform its monitoring functions.
- Evidence: Multiple steps in SKILL.md provide CLI fallbacks using the
caveman cloudcommand structure (e.g.,caveman cloud whoami,caveman cloud traces show).
Audit Metadata