No documented output schemas for any tool. LLMs cannot infer what fields are returned, forcing them to guess at response structure and breaking tool composition chains.

Private key parameters accept both hex and nsec formats, but no validation is visible and no error guidance for invalid formats. LLMs may pass malformed keys without clear feedback on what went wrong.

createKeypair has an enum 'format' parameter but the schema shows type 'enum' without listing the actual enum values. LLMs cannot know what formats are supported.

convertNip19 has targetType as enum with no visible values in schema. Requires relays and author/kind/identifier for different entity types, but these conditional dependencies are not documented in parameter descriptions.

No error classification or recovery guidance visible in tool definitions. When operations fail (e.g., relay unreachable, invalid signature), there is no guidance on whether to retry, ask the user, or abort.

Many parameters lack detail about expected format or constraints. E.g., 'relays' parameter description says 'Optional list of relays to query' but does not specify the URL format, protocol (wss://?), or what happens if a relay is unreachable.

Compound operations like 'createNote + signNote + publishNote' are offered as separate tools. While this is good for composability, there is no guidance on when to call them together or what intermediate state looks like. The helper functions (formatProfile, formatNote) suggest response structure but are not exposed in tool definitions.

postAnonymousNote generates a one-time keypair and publishes anonymously. The description mentions this but does not warn that the keypair is not persisted, if the agent needs to retrieve or reply from this note, it will not have the private key.