Neynar Inbox

The skill's stated purpose (email API for agents) matches the runtime instructions (endpoints to create mailboxes, send/receive email). However there is an inconsistency between the declared homepage (https://email.neynar.ai in metadata) and the actual API base URL used in the instructions (https://neynar-inbox-api.rish-68c.workers.dev). The SKILL.md also exposes behavior (create mailbox without auth, returns apiKey) that implies the agent must manage secrets even though the skill metadata lists no primary credential — this mismatch is concerning.

Instructions are explicit about API calls and do not instruct reading arbitrary local files or unrelated credentials. They do, however, instruct the agent to store the returned apiKey securely and recommend polling every 30–60 seconds or registering webhooks — both of which imply persistent secret storage and regular network activity outside the agent's immediate interaction. That scope expansion (scheduling/persisting keys) is not declared elsewhere.

This is an instruction-only skill with no install spec or code files, which is lower risk. One minor inconsistency: the SKILL.md lists requires.bins: [curl], while the registry metadata provided earlier listed no required binaries. No downloads or extractable installs are present.

The skill will produce and require an API key for authenticated calls, but requires.env and primary credential fields are empty in the package metadata. The runtime examples show Authorization: Bearer <apiKey>, so the agent or user will need to store that secret somewhere — yet the skill does not declare or request that storage. Absence of declared credential handling is a mismatch and could lead to accidental secret leakage if the agent stores the key insecurely.

always is false (good). The skill recommends polling every 30–60s or registering webhooks, and the platform allows autonomous invocation by default. That combination gives an agent the ability to send/receive arbitrary emails and perform frequent outbound connections; this is expected for an email integration but increases the blast radius if the agent acts without user oversight. Consider limiting autonomous use or monitoring outbound messages.