The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates reading untrusted data from the Ethereum blockchain, which serves as a surface for indirect prompt injection attacks.
Ingestion points: Untrusted data enters the agent context through methods such as provider.getBlock, provider.getTransaction, and provider.lookupAddress in examples/read-chain-data/README.md, as well as token.queryFilter and event listeners in examples/listen-events/README.md.
Boundary markers: The skill does not implement boundary markers or instructions to isolate retrieved blockchain content from the agent's primary instructions.
Capability inventory: The skill provides capabilities to execute on-chain transactions and approve contract interactions, as documented in examples/send-transaction/README.md, examples/interact-contract/README.md, and templates/ethers-client.ts.
Sanitization: No sanitization or validation routines are included to process external blockchain data before it is returned to the agent.

ethers-js