strategy-discovery

Pass

Audited by Gen Agent Trust Hub on Nov 21, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external data from local backtest runs and manifests which serves as a potential vulnerability surface.
  • Ingestion points: The refresh_strategy_evidence tool reads data from JSON manifests and individual run artifacts such as trades.csv and equity.csv.
  • Boundary markers: The system enforces strict path-containment using an allowed roots list (via VIBE_TRADING_ALLOWED_RUN_ROOTS) to prevent directory traversal or unauthorized file system access.
  • Capability inventory: The skill is capable of reading local CSV and JSON files and writing results to a local, disposable SQLite database used as an evidence cache.
  • Sanitization: It implements a 'Hard-Gate Checklist' that verifies run success, metrics existence, and ensures numerical data is finite (no NaN values) before any data is committed to the cache.
  • [COMMAND_EXECUTION]: The skill documentation and tools rely on an existing local command-line utility (vibe-trading) to perform evidence refreshes and strategy adaptations, following standard local tool integration patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 21, 2026, 05:36 PM