The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains a hardcoded secret_key in references/openapi/_shared.md and scripts/gate-api-call.py. While described as public defaults for a basic tier, hardcoding functional secret keys for a financial API is a significant security risk.\n- [DATA_EXFILTRATION]: Instructions in references/openapi.md and references/openapi/sign.md explicitly prompt users to paste their private keys or mnemonics into the chat. Chat platform backends frequently log conversation history, meaning this practice exposes raw private keys to third-party services and potential log breaches.\n- [DYNAMIC_EXECUTION]: The skill guides the agent to generate and execute dynamic code blocks (e.g., via node -e) for signing SUI and TON transactions as seen in references/openapi/sign.md. This runtime code generation for handling cryptographic operations increases the risk of injection or logic errors.\n- [COMMAND_EXECUTION]: The skill relies on the execution of multiple local scripts including gate-api-call.py, sign-tx-evm.py, and sign-tx-sol.js to perform its core operations.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes various strings from external API responses (such as token symbols and route information) without explicit sanitization or boundary markers, creating a surface for indirect prompt injection attacks via malicious API data.

gate-dex-trade