The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and process external data from the Treasures API, including stock tickers, price snapshots, and portfolio holdings. This data is used to drive critical logic such as trade quoting, route selection, and execution. The instructions lack explicit prompt boundary markers or sanitization requirements for this externally sourced content, which constitutes an attack surface for indirect prompt injection where malicious API responses could influence agent behavior.
Ingestion points: Data is retrieved from the /stocks/tickers, /stocks/prices, /portfolio, and /trades endpoints (documented in references/data.md).
Capability inventory: The agent is empowered to sign and submit blockchain transactions, set ERC-20 allowances, and execute cross-chain bridges (references/trading.md, references/bridging.md).
Boundary markers: No delimiters or instructions are provided to the agent to treat API responses as untrusted data.
Sanitization: No validation or escaping of API-returned strings is required before they are interpolated into subsequent agent context or used in decision-making.
[CREDENTIALS_UNSAFE]: The references/auth.md file contains a hardcoded private key seed (0x0102030405060708090a0b0c0d0e0f101112131415161718191a1b1c1d1e1f20). However, this is presented strictly as a deterministic test vector for developers to verify their signing logic implementations and is not an active credential for a production account.

treasures-b2b-api