The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill utilizes WebFetch and the Upstash Context7 MCP server to retrieve external documentation. While this introduces a potential ingestion point for untrusted data, the skill includes a specific safety guardrail (Rule 11) instructing the agent to treat this data as untrusted and ignore any malicious instructions contained within.
[CREDENTIALS_SAFE]: The skill implements strict policies regarding sensitive information. It explicitly forbids reading, printing, or hardcoding API secrets, private keys, or wallet material, and advises users to keep credentials in environment variables or backend processes.
[COMMAND_EXECUTION]: The skill establishes a 'Non-Custodial Action Protocol' and an 'Execution Boundary' that prevents the agent from executing state-changing transactions, deploying contracts, or signing data. It instead provides the user with command templates or preflight packages to review and execute in their own controlled environment.

chainlink-data-streams-skill