interpret-feedback

Pass

Audited by Gen Agent Trust Hub on Apr 18, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill processes third-party feedback from external URLs or files, creating an indirect prompt injection surface. The ingestion of untrusted data combined with the capability to spawn subagents creates a pathway for content-driven manipulation.\n
  • Ingestion points: Step 1 involves reading file paths or fetching content from URLs provided in the conversation context (SKILL.md).\n
  • Boundary markers: The instructions do not define boundary markers (such as XML tags or specific delimiters) to isolate the untrusted feedback content from the agent's logic.\n
  • Capability inventory: The skill can read local files, fetch remote content via URLs, and spawn subagents with tool-use capabilities using the opus model (SKILL.md).\n
  • Sanitization: No sanitization or validation of the ingested content is performed before it is processed by the interpretation subagents.\n- [EXTERNAL_DOWNLOADS]: Step 1 permits the agent to fetch content from user-supplied URLs to identify feedback items. While this is necessary for the skill's function, it involves interaction with potentially untrusted external domains.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 18, 2026, 11:00 PM