caveman-explore

Pass

Audited by Gen Agent Trust Hub on Aug 11, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill is strictly limited to read-only operations using Glob, Grep, and Read tools. No capabilities for file modification, command execution, or network communication are present, aligning with the stated purpose of a repository explorer.- [PROMPT_INJECTION]: The skill functions by reading untrusted data from a repository, creating a surface for indirect prompt injection. However, the risk is mitigated by the skill's design.
  • Ingestion points: File content retrieved via Read and Grep tools (SKILL.md).
  • Boundary markers: Not explicitly defined for tool output, but the agent is instructed to focus on compact citations.
  • Capability inventory: Highly restricted to read-only search tools; no execution or exfiltration capabilities (SKILL.md).
  • Sanitization: The output format is strictly limited to file paths and line ranges, which naturally prevents the agent from echoing or following malicious instructions found within the repository files.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 11, 2026, 02:46 PM
Security Audit — agent-trust-hub — caveman-explore