The Agent Skills Directory

[SAFE]: The skill performs read-only network operations to Intuition's official GraphQL endpoints for identity preflight and registry verification. These operations target well-known vendor infrastructure and use neutral fetching logic.
[SAFE]: The implementation includes advanced security-positive features, such as the isPrivateHostname function in scripts/partner-plan.mjs, which effectively prevents Server-Side Request Forgery (SSRF) by blocking access to local, private, or loopback IP ranges during URL validation.
[SAFE]: Data integrity is maintained through SHA-256 hashing. The skill verifies the integrity of the bundled registry and the generated semantic plans, ensuring that any manual tampering or corruption is detected before the plan can be used in later phases.
[SAFE]: The skill follows a strict 'approval ladder' that enforces human-in-the-loop boundaries. It explicitly instructs the agent to stop after generating a plan and prohibits autonomous transaction execution, pinning, or broadcasting in a single turn.
[PROMPT_INJECTION]: Instructions in SKILL.md contain robust defensive directives, commanding the agent to treat all external data (manifests, registration files, and resolver responses) as untrusted and to ignore any embedded instructions that attempt to bypass preflight or safety protocols.

erc8004-agent-layer