partykit

Pass

Audited by Gen Agent Trust Hub on Mar 29, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill demonstrates an implementation of AI functionality that is susceptible to indirect prompt injection. Raw strings from user messages are directly interpolated into model prompts.
  • Ingestion points: Data enters the system via the onMessage handler from a Party.Connection in SKILL.md.
  • Boundary markers: There are no delimiters (like XML tags or triple quotes) or system instructions used to separate the user-provided text from the model prompt.
  • Capability inventory: The skill uses the partykit-ai package to run inference on the @cf/meta/llama-3-8b-instruct model and can broadcast the resulting AI response to all connected users.
  • Sanitization: No validation, escaping, or filtering of the data.text variable is performed before it is sent to the AI service.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 29, 2026, 07:53 AM