The Agent Skills Directory

[COMMAND_EXECUTION]: Instructions provide example shell scripts utilizing standard utilities including curl, jq, openssl, and cast for signing messages and interacting with the Berry API.- [DATA_EXFILTRATION]: The skill performs network operations to juicerapi.berryfi.org and api.bankr.bot. These domains correspond to the official Berry Finance service and the author's (BankrBot) API infrastructure, representing legitimate operational traffic for authentication and inference.- [INDIRECT_PROMPT_INJECTION]: The skill handles untrusted data from API responses and model outputs.
Ingestion points: Data enters the context via Berry API response fields (e.g., token metadata, balances) and external model content.
Boundary markers: The skill contains a dedicated 'Untrusted content and prompt-injection boundary' section that explicitly instructs the agent to treat external content as data only and never as operational instructions.
Capability inventory: Capabilities include wallet signing via api.bankr.bot and API interactions via curl to juicerapi.berryfi.org.
Sanitization: The skill mandates a fail-closed verification process where all transaction parameters (chainId, contract addresses, amounts) must be verified against human input or pinned allowlists, effectively neutralizing instructions embedded in untrusted data.

berry-juicer