skills/rheinmir/setup/caveman/Gen Agent Trust Hub

caveman

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill instructs the agent to adopt a 'caveman' persona. This is a stylistic constraint for token efficiency and does not attempt to bypass core safety or ethical guidelines.
  • [INDIRECT_PROMPT_INJECTION]: The 'Output Report' feature creates markdown files and updates indices based on user-supplied summaries and task outputs. This configuration presents a surface for indirect injection.
  • Ingestion points: The user-provided summary (the variable) and the task results are interpolated into file paths and document bodies (SKILL.md).
  • Boundary markers: The draft template lacks explicit delimiters or instructions to treat user-provided content as untrusted.
  • Capability inventory: The skill utilizes file system write capabilities to create drafts and append data to index.md and log.md in the llmwiki/ directory.
  • Sanitization: No sanitization or validation logic is specified for the summary or the reported output, which could be exploited to attempt directory traversal or corrupt the wiki structure.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 07:41 PM
Security Audit — agent-trust-hub — caveman