The Agent Skills Directory

[SAFE]: The skill establishes a multi-layered validation logic for financial transactions. Section 0 defines hardcoded 'desk ceilings' (e.g., max risk, max leverage, daily loss stops) that act as a fundamental security boundary. These ceilings are designed to override any potentially corrupted or malicious user-provided configuration files, ensuring the agent cannot authorize actions beyond safe desk-wide limits.
[SAFE]: Data ingestion is restricted to specific local workspace paths (/workspace/trading-desk/risk-limits.md) and official account state data fetched through referenced skills (hyperliquid-account, hyperliquid-market-data). This controlled interface minimizes the risk of processing untrusted external data that could lead to indirect prompt injection.
[SAFE]: The skill includes detailed arithmetic instructions for trade sizing, specifically using 'stressed distance' calculations to account for slippage and fees. This prevents unintentional risk escalation that could occur if nominal stop distances were used, representing a safety-first approach to automated trading logic.
[SAFE]: All identified sensitive operations, such as accessing account balances and issuing trade tickets, are documented as part of the intended functional scope for a risk management tool. There are no indications of data exfiltration to non-vendor domains or unauthorized persistence mechanisms.

desk-risk-limits