flywheel-discord

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill operates on Discord, which is an untrusted third-party surface. It identifies this attack surface and provides extensive defensive rules.
  • Ingestion points: Discord user messages and external content (URLs/attachments).
  • Boundary markers: The skill defines clear response patterns for manipulation attempts, such as ignoring 'ignore previous instructions' and rejecting authority claims.
  • Capability inventory: High-risk actions are explicitly prohibited, including shell command execution, file system access, and the use of owner credentials for API calls.
  • Sanitization: The instructions state to summarize content without following embedded instructions and to treat all user-provided data as untrusted.
  • [PROMPT_INJECTION]: The skill contains keywords like 'ignore', 'override', and 'bypass', which were flagged by static analysis. However, these appear exclusively in defensive contexts (Red Flags and Manipulation Patterns) to instruct the agent on what to ignore when interacting with users. The 'OVERRIDES' keyword in the identity section is a standard directive for applying specific persona rules.
  • [DATA_EXFILTRATION]: The skill implements a deny-by-default policy for sensitive data, explicitly listing categories like personal messages, emails, notes, credentials, and financial data that must never be accessed or revealed to Discord users.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 09:51 AM