The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically imports modules from a path provided via the --circle-cli-repo command-line argument. The loadCircleRuntime function in scripts/_shared.mjs uses import() to load TypeScript modules and libraries (such as viem) from this computed path, allowing the execution of code from an unverified external location.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources to construct sensitive blockchain transactions.\n
Ingestion points: Blockchain logs and transaction receipts are fetched via RPC, and intent status data is retrieved from the Eco API in scripts/eco-self-refund.mjs.\n
Boundary markers: While the internal logic performs rigorous validation (e.g., recomputing intent hashes and simulating transactions), there are no explicit prompt-level delimiters to separate this untrusted data from agent instructions.\n
Capability inventory: The skill has the capability to submit transactions via the Circle Agent Wallet API and write to the local file system.\n
Sanitization: The sanitize and safeErrorMessage functions are employed to redact secrets, tokens, and raw signatures from all logs and evidence files.\n- [EXTERNAL_DOWNLOADS]: The skill makes network requests to well-known service endpoints including api.eco.com, quotes.eco.com, and blockchain RPC providers to retrieve data necessary for its primary fund recovery functions.

recover-eco-funds