The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell-based tools including curl, jq, cast, and open/xdg-open. It mitigates command injection risks through mandatory single-quoting of user-provided values and rigorous variable expansion quoting in shell commands. It specifically restricts tool usage to a defined set of trusted crypto-related domains and RPC endpoints.
[EXTERNAL_DOWNLOADS]: Market information and token metadata are retrieved from official PancakeSwap repositories and well-known industry services such as DexScreener, DefiLlama, and CoinGecko. These downloads are restricted to informational data and do not involve executable scripts or packages.
[PROMPT_INJECTION]: The skill incorporates a comprehensive defense against indirect prompt injection. 1. Ingestion points: External data enters via token lists and market APIs described in SKILL.md and data-providers.md. 2. Boundary markers: Explicit 'MANDATORY SECURITY RULES' define the boundary between system instructions and external data. 3. Capability inventory: Commands like curl and cast are available but strictly scoped. 4. Sanitization: The skill mandates regex-based address validation and instructs the agent to treat API fields as untrusted text rather than executable instructions.

liquidity-planner