query

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill searches local directories (.fpf/knowledge/, .fpf/decisions/, .fpf/evidence/) and displays the contents of matching files to the user. If an attacker can inject malicious natural language instructions into these files (e.g., via a pull request or untrusted database dump containing hypotheses), the agent may execute or adopt those instructions when the skill parses and presents them.
  • Ingestion points: Reading text and files under .fpf/knowledge/, .fpf/decisions/, and .fpf/evidence/ subdirectories.
  • Boundary markers: Absent. No special delimiters or explicit rules are defined to tell the agent to treat file contents purely as untrusted data.
  • Capability inventory: The skill lists file display and table presentation capabilities; no active network, subprocess, or execution functions are defined within this markdown specification.
  • Sanitization: Absent. The file contents are displayed directly in table and markdown text formats.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 01:09 AM
Security Audit — agent-trust-hub — query