wolf-howl

Pass

Audited by Gen Agent Trust Hub on Feb 27, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill implements a sub-agent workflow that processes external and historical trade data. This creates a surface for indirect prompt injection where data ingested from external APIs or logs could attempt to influence the agent's analysis or reporting behavior.
  • Ingestion points: Data enters the context through 'memory/' logs, local DSL state JSON files, and trade history queried from the Senpi MCP connection via 'mcporter'.
  • Boundary markers: The sub-agent instructions in 'references/analysis-prompt.md' do not define clear delimiters or separators to isolate untrusted trade data from the core analysis instructions.
  • Capability inventory: The sub-agent is capable of reading workspace files, writing new report files to the filesystem, and sending messages to a user-provided Telegram chat ID.
  • Sanitization: The skill does not perform explicit sanitization or structural validation on the external trade data before it is interpolated into the analysis prompt.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 27, 2026, 09:43 PM