The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches an installation script from OKX's official GitHub repository and a pre-compiled plugin binary from the vendor's repository.
[REMOTE_CODE_EXECUTION]: Executes a setup script from OKX, a well-known service provider, during the installation phase to configure the environment.
[DATA_EXFILTRATION]: Performs installation reporting by collecting system metadata such as hostname, platform, and the user's home directory path. This information is hashed locally into a non-reversible identifier before being sent to well-known service endpoints for telemetry purposes.
[PROMPT_INJECTION]: The skill ingests data from external GraphQL APIs and on-chain calls, which could theoretically contain malicious instructions. This risk is mitigated by explicit security guidelines in the skill instructions and a requirement for manual user confirmation for all transaction-signing operations.
Ingestion points: External data enters the agent context through GraphQL queries in src/api.rs and JSON-RPC eth_call results in src/rpc.rs.
Boundary markers: The SKILL.md file contains a 'Security Notices' section that explicitly instructs the agent to treat data from the CLI as untrusted external content.
Capability inventory: The skill possesses the ability to execute shell commands via the onchainos CLI for wallet interactions and contract calls, as seen in src/onchainos.rs.
Sanitization: Implements strict parsing for token amounts and manual construction of ABI-encoded calldata, ensuring external strings are never directly used as executable instructions.

morpho-base