The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the XRPL blockchain, creating a potential surface for indirect prompt injection attacks.
Ingestion points: The skill ingests external data by reading order book entries, account offers, and token/issuer metadata through book_offers, account_info, and account_offers RPC requests (located in SKILL.md and references/trading.md).
Boundary markers: The skill implements a mandatory pre-trade summary step (Step 4 in SKILL.md) that requires the user to acknowledge the specific assets, network, and estimated slippage before any transaction is prepared.
Capability inventory: The skill possesses the capability to construct OfferCreate and OfferCancel transactions, although it delegates the final signing and submission to an external XRPL Agent Wallet skill (SKILL.md Flows 1 and 2).
Sanitization: The instructions include a dedicated "Untrusted input" section that explicitly warns the agent to treat ledger data strictly as data and not as instructions, highlighting the risk of crafted strings in transaction results or lookalike issuer addresses.

xrpl-trading