The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands (cast receipt, cast tx, cast run) that incorporate a {RPC_URL} parameter. The instructions explicitly state that users may provide a custom RPC URL. While the skill mandates a strict regex validation (^0x[a-fA-F0-9]{64}$) for the transaction hash, it does not specify any validation or sanitization for the RPC URL string, which could be used as a vector for command injection if the agent executes these commands via a shell without proper argument escaping.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted data retrieved from blockchain RPC nodes to diagnose transaction failures.
Ingestion points: Output from cast commands and ethers.js calls, including transaction receipts, logs, and revert reasons, are ingested into the agent context in SKILL.md (Steps 3, 4, and 5).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the output templates used to display transaction results.
Capability inventory: The skill has the capability to execute shell commands (cast) and perform network requests (via RPC providers).
Sanitization: Input validation is applied to the transaction hash via regex, but no sanitization is mentioned for the data returned from external RPC providers before it is parsed for revert reasons or displayed to the user.

swap-status