The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The Python snippet in SKILL.md contains a load_key function that attempts to read a sensitive cryptographic private key from a local file path (~/.hyperliquid/api-wallet.key) if the corresponding environment variable is not set. Accessing hardcoded sensitive file paths in the user's home directory for secrets is a credential exposure risk.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, including exchange metadata, market prices, and order execution responses, which are then used to influence agent decisions regarding order placement and reconciliation. This creates a vulnerability surface where malformed or malicious data from the external API could potentially manipulate the agent's logic.
Ingestion points: Data is ingested from info.meta(), info.all_mids(), info.user_state(), and the results of exchange.order() and exchange.bulk_orders() calls in both Python and TypeScript snippets.
Boundary markers: The skill lacks explicit delimiters or specific instructions to distinguish between trusted local context and external API data, increasing the risk of the agent obeying instructions embedded in API responses.
Capability inventory: The skill possesses significant financial capabilities, including placing, cancelling, and modifying orders on a live trading exchange via the Exchange and ExchangeClient classes.
Sanitization: While the skill implements price and size rounding logic, it lacks validation or sanitization of the logical content contained within API responses before they are processed by the agent for subsequent actions.

hyperliquid-orders