The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes live transaction data from blockchain mempools, which represents a surface for indirect prompt injection if an attacker crafts transaction fields (like string parameters in a contract call) containing natural language instructions. However, the risk is mitigated by the tool's focus on structured decoding.
Ingestion points: rpc_client.py fetches raw transaction data via JSON-RPC calls (txpool_content, eth_pendingTransactions).
Boundary markers: Data is converted into structured Python objects (PendingTransaction, DecodedCall) and presented via formatted tables or JSON.
Capability inventory: The skill uses the requests library for network communication and has access to Bash, Read, and Write tools.
Sanitization: The TransactionDecoder in tx_decoder.py explicitly maps hex data to known method signatures and structured parameters, reducing the likelihood that the agent interprets data as instruction.
[EXTERNAL_DOWNLOADS]: The skill connects to well-known, public Ethereum-compatible RPC endpoints (e.g., eth.llamarpc.com, polygon-rpc.com, arb1.arbitrum.io) to retrieve mempool and block data. These connections are standard for blockchain monitoring tools and target established infrastructure providers.

analyzing-mempool