grilling

Pass

Audited by Gen Agent Trust Hub on Aug 19, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill instructions mandate the use of sub-agents to autonomously retrieve "facts" from the environment (filesystem, tools) to answer questions about user-provided plans or ideas. This creates an attack surface where instructions embedded in the data being analyzed could be executed or obeyed by the agent or its sub-agents.\n
  • Ingestion points: The agent ingests user-provided plans, decisions, and ideas as the basis for the design tree, as well as answers provided during the "rounds" of interviewing (SKILL.md).\n
  • Boundary markers: The instructions lack boundary markers or warnings to the agent to disregard instructions contained within the user's plan or the retrieved environment facts.\n
  • Capability inventory: The skill explicitly grants the capability to access the filesystem and execute tools through sub-agents to satisfy fact-finding requirements (SKILL.md).\n
  • Sanitization: No sanitization, validation, or filtering of the user input or retrieved environment data is specified.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 19, 2026, 05:19 PM
Security Audit — agent-trust-hub — grilling