The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents a significant attack surface for indirect prompt injection by combining powerful financial and system capabilities with the ingestion of untrusted external data.
Ingestion points: The agent ingests data from user-supplied prompts, social media handle resolution (Twitter, Farcaster), and external web content through a built-in headless browser.
Boundary markers: The documentation does not describe explicit use of prompt delimiters or specific instructions to ignore embedded commands when processing content from the headless browser or social metadata.
Capability inventory: The skill possesses extensive capabilities across multiple chains, including token swaps, multi-recipient transfers, raw transaction submission with arbitrary calldata, and persistent file storage operations.
Sanitization: Safety controls are implemented at the wallet level, including default $500 daily and per-transaction spending limits, a 15% price impact limit, and mandatory location verification for trading tokenized equities.
[DYNAMIC_EXECUTION]: The skill enables the agent to generate and deploy executable code at runtime.
The x402 Cloud feature allows the agent to write handler code for paid API endpoints and deploy them directly to the x402 cloud infrastructure.
This involves the dynamic generation of script content based on user requirements and its subsequent execution or deployment in a remote environment.

bankr