risk-manage

Pass

Audited by Gen Agent Trust Hub on Aug 8, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill uses authoritative instructions such as 'single source of truth' and 'agent is not permitted to override its outputs' to influence agent behavior and potentially bypass other system constraints.
  • [PROMPT_INJECTION]: The skill processes untrusted data from multiple files (signals.json, positions.json, forecasts.json, markets.json) to make automated financial decisions, creating a surface for indirect prompt injection. 1. Ingestion points: runtime/state/positions.json, runtime/state/signals.json, runtime/state/forecasts.json, runtime/state/markets.json. 2. Boundary markers: Absent. 3. Capability inventory: Filesystem writes, database updates, and network webhook heartbeats. 4. Sanitization: Not explicitly implemented in the skill instructions; relies on external library 'weather_runtime'.
  • [DATA_EXFILTRATION]: The skill transmits status information ('heartbeats' and 'close-all summary') to an external webhook, representing a potential automated data egress channel for portfolio state.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 8, 2026, 11:46 AM
Security Audit — agent-trust-hub — risk-manage