The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download and execute installation scripts from the vendor's official domain (goldsky.com) and a well-known industry service (foundry.paradigm.xyz). These scripts are used to install the necessary command-line tools for the development environment.
[COMMAND_EXECUTION]: The skill uses local shell commands for project scaffolding, directory management, and build cache cleanup (e.g., rm -rf .compose/). These operations are routine for software development tasks.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data from blockchain events which could be manipulated by an attacker to attempt to influence agent behavior.
Ingestion points: Untrusted data enters the agent context through the payload: OnchainEvent argument in src/tasks/on-transfer-requested.ts.
Boundary markers: The skill uses structured decoding via evm.decodeEventLog with a predefined ABI, which serves as a validation boundary for the incoming event data.
Capability inventory: The skill has the capability to execute blockchain transactions through wallet.writeContract as seen in src/tasks/on-transfer-requested.ts.
Sanitization: Before performing sensitive operations, the skill passes the external data to a compliance screening service (screenWallet) to validate the reputation and status of the addresses involved.
[SAFE]: The skill enforces strong credential safety policies, including instructions to use gas-sponsored smart wallets that do not require private key management by the user, and explicit requirements to manage API keys as environment secrets without exposing them in the agent's context or history.

compose-compliance-oracle