The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is implemented as a CLI tool intended to be executed via the bun runtime (inbox/inbox.ts).
[EXTERNAL_DOWNLOADS]: The skill communicates with external APIs, specifically aibtc.com for inbox operations and Stacks blockchain infrastructure (Hiro API) for transaction and account status.
[CREDENTIALS_UNSAFE]: The send subcommand utilizes getAccount() to access the wallet's private key for signing sBTC transfer transactions. While this is a functional requirement for sending paid messages, it involves handling highly sensitive credentials in memory.
[INDIRECT_PROMPT_INJECTION]: The read command ingests message content from external, potentially untrusted sources via the inbox API.
Ingestion points: Data enters the agent context through the read command which fetches messages from https://aibtc.com/api/inbox/{address}.
Boundary markers: No explicit boundary markers or safety instructions are applied to the message content in the provided code.
Capability inventory: The skill possesses transaction signing (makeContractCall) and network communication (fetch) capabilities.
Sanitization: The implementation lacks sanitization or filtering of the content field from received messages, allowing potential malicious instructions to reach the agent.
[DATA_EXFILTRATION]: The skill transmits message content and serialized transaction hexes to the AIBTC API to complete the messaging protocol. While these transmissions are part of the intended service, they involve sending agent-controlled data to an external platform.

inbox