The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection vulnerability surface.
Ingestion points: Data is fetched from the blockchain via the BMAP API in scripts including read-posts.ts, read-messages.ts, read-follows.ts, read-friends.ts, and read-likes.ts.
Boundary markers: The scripts do not use delimiters or warning markers when displaying retrieved content such as post content or message bodies to the agent.
Capability inventory: The skill has the ability to broadcast transactions to the network and execute shell commands via the Bun runtime.
Sanitization: No validation or sanitization is performed on the content retrieved from the BMAP API before it is presented to the agent, allowing for potential injection of instructions via on-chain data.
[COMMAND_EXECUTION]: Handling of sensitive private keys.
The skill documentation and creation scripts (e.g., create-post.ts, create-like.ts, create-follow.ts) require a Wallet Import Format (WIF) private key as a command-line argument.
While this is the intended mechanism for signing on-chain transactions, providing private keys to an automated agent environment involves risk if the session or environment is compromised.

bsocial