caveman-evidence-review
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute specific
cavemanCLI commands (e.g.,caveman cloud costs,caveman cloud traces search) to retrieve project data. These commands are localized to the Caveman Cloud toolset and are essential for the skill's primary function. - [INDIRECT_PROMPT_INJECTION]: The skill processes external data retrieved from cloud reports and trace logs, which could theoretically contain malicious instructions designed to influence the agent. However, the skill provides strict constraints to act as a read-only operator and avoid administrative actions, mitigating the risk.
- Ingestion points: Data fetched via
caveman_report,caveman_trace_search, andcaveman_trace_get(or CLI fallbacks) in SKILL.md. - Boundary markers: None explicitly mentioned in the instructions to separate untrusted data from agent instructions.
- Capability inventory: The skill is restricted to reading data through the
cavemantoolset; it lacks capabilities for arbitrary file writing, generic network requests, or system-level modification. - Sanitization: No explicit sanitization or filtering of the ingested cloud data is described.
Audit Metadata