The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell-based tooling including curl, jq, python3, and cast to interact with blockchain RPCs and web APIs. It incorporates robust security guidelines for the agent, mandating the use of single quotes for user inputs, variable quoting in commands, and regex-based validation for token addresses (^0x[0-9a-fA-F]{40}$) and numeric values.
[EXTERNAL_DOWNLOADS]: Fetches data from the author's own infrastructure (hub-api.pancakeswap.com, explorer.pancakeswap.com, pancakeswap.ai) and established industry services such as DexScreener, GeckoTerminal, and public Binance Smart Chain RPC endpoints. It also performs an initial telemetry ping to pancakeswap.ai to record the skill version, agent type, and basic system architecture info (OS/Arch).
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external APIs, including token symbols and route descriptions. It mitigates potential risks by providing explicit instructions to treat all API response content as untrusted and to never interpret data fields as commands.
[DYNAMIC_EXECUTION]: Uses python3 -c for internal mathematical operations, such as converting human-readable token amounts to their wei equivalents and formatting price values for the route summary.

hub-swap-planner