senpi-trader-research

Pass

Audited by Gen Agent Trust Hub on Aug 21, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructions require the agent to execute a local Python script (scripts/research.py) using the host environment to perform trader discovery and analysis.
  • [EXTERNAL_DOWNLOADS]: The skill's communication helper (scripts/mcp_client.py) performs network operations to mcp.prod.senpi.ai, which is the official API endpoint for the vendor ('senpi-ai'). These operations are limited to fetching market data and trader statistics.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external data originating from the Hyperliquid exchange, creating a potential surface for indirect prompt injection if trader metadata contains malicious instructions.
  • Ingestion points: Data is ingested via discovery_ and leaderboard_ API calls performed in scripts/research.py.
  • Boundary markers: The SKILL.md file provides clear instructions to "Read its JSON" and only use values returned by the engine, establishing logical boundaries for the agent's interpretation.
  • Capability inventory: The skill has the capability to execute local scripts and perform network communication with the vendor's API.
  • Sanitization: External data is strictly parsed as JSON using Python's standard library before processing, minimizing structural confusion.
  • [DYNAMIC_EXECUTION]: The main engine script dynamically imports its helper module (mcp_client) from the local skill directory using standard Python import mechanisms.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 21, 2026, 10:31 PM
Security Audit — agent-trust-hub — senpi-trader-research