The Agent Skills Directory

[SAFE]: The skill implements strong security practices and defensive instructions.
[CREDENTIALS_SAFE]: The 'Safety Guardrails' explicitly forbid the agent from reading, opening, or echoing local wallet credentials, signing materials, keystores, or secret environment files (e.g., those containing PRIVATE_KEY or RPC_URL). It also prohibits asking the user to provide such secrets.
[INDIRECT_PROMPT_INJECTION]: The skill includes a specific guardrail (Rule 11) instructing the agent to treat all external documentation, repository content, and API responses as untrusted data. It explicitly directs the agent to ignore any instructions found in those external sources that attempt to trigger credential access, shell execution, or bypass safety rules.
[EXTERNAL_DOWNLOADS]: The skill retrieves documentation and source code from official Chainlink sources, including the smartcontractkit/chainlink-ace GitHub repository and docs.chain.link. These are well-known and trusted services used for reference and guidance.
[COMMAND_EXECUTION]: While the skill permits the use of the Bash tool, it limits this to read-only work, explanations, and local test planning. It requires a rigorous 'Approval Protocol' (a detailed preflight summary and explicit user confirmation) before any action that would modify on-chain state or configuration.

chainlink-ace-skill