skills/pandoscope/skills/take-probe/Gen Agent Trust Hub

take-probe

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill loads and follows instructions from an external file (probe.md) located in a directory provided via user message.\n
  • Ingestion point: Reads <ref>/probe.md based on a reference from a message (SKILL.md).\n
  • Boundary markers: Absent; instructions specify following the file 'verbatim'.\n
  • Capability inventory: The agent follows instructions which may use any available tools and executes check.sh (SKILL.md).\n
  • Sanitization: Absent; no filtering or validation of the probe content is described.\n- [COMMAND_EXECUTION]: The skill executes a bundled shell script check.sh for validation.\n
  • Evidence: 'Run check.sh beside this file.' (SKILL.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 06:03 PM
Security Audit — agent-trust-hub — take-probe