caveman
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to apply a terse formatting style to all outputs, which may include data derived from untrusted external sources.
- Ingestion points: The agent processes user queries and potential tool outputs (e.g., file reads, terminal output) described in the SKILL.md rules.
- Boundary markers: No specific delimiters or boundary markers are defined to isolate untrusted data within the instructions.
- Capability inventory: The skill is intended for use in environments like Claude Code, which have filesystem and terminal access.
- Sanitization: The risk is mitigated by Rule 6 (Payload verbatim), which protects technical data from being mangled, and mandatory exceptions that force the agent to revert to standard prose for security warnings and irreversible actions.
- [SAFE]: No evidence of malicious behavior, credential exfiltration, obfuscation, or unauthorized system access was found. The skill's instructions for configuration management (e.g., ~/.config/caveman/config.json) follow standard software practices.
Audit Metadata