wechat-fastbridge demonstrates solid definition quality with explicit tool registration, complete input schemas using Zod, and good parameter documentation. All 6 tools are clearly defined with verb-noun naming (wechat_status, wechat_read, wechat_send, wechat_send_media, wechat_wait, wechat_inbox_wait). Input schemas are well-constrained with type definitions, min/max bounds, enums, and defaults. However, descriptions are notably brief (10-60 chars), falling short of the 50-200 char LLM-optimized baseline. Output schemas are not explicitly documented, responses are shaped by the `reply()` and `compact()` functions, but the structure is not formally declared in tool definitions. Error handling exists but is minimal: the `exclusive()` wrapper catches errors and calls `errorReply()`, which returns a fixed error structure, but lacks actionable recovery guidance or categorization (retryable vs fatal). The `wechat_send_media` tool includes inline validation for conditional parameters (FILE_PATH_REQUIRED, STICKER_QUERY_REQUIRED), which is good, but other tools lack similar guardrails. No tool annotations for destructiveness or idempotency beyond read-only hints. No pagination support despite tools that might return variable-length data (wechat_read, wechat_wait context arrays). Security is strong: no API keys in parameters, native macOS-only execution prevents unauthorized access, and the `exclusive()` mutex ensures thread-safe serial execution. Tool composition is well-designed, each tool has a single responsibility, and tools chain via shared chat/signature identifiers.

• Expand tool descriptions to 80 - 150 chars each. Example for wechat_read: 'Read messages from a chat (up to 20). Optionally fetch messages after a known signature for delta sync, or include smart contextual enrichment (0-4 levels). Use this for polling; use wechat_wait() to block until changes arrive.'
• Document output schemas explicitly in tool definitions. Use structuredContent or a new responseSchema field. Example: 'Returns {ok: bool, chat: string, messages: [{id, text, sender, timestamp}], context: [string], signature: string, ms: number}.'
• Enhance error responses with recovery hints. When kind='file' and path is missing, return 'FILE_PATH_REQUIRED: Include path parameter when kind=file. Example: {kind: "file", path: "/path/to/doc.pdf", ...}'. When wechat_wait times out, clarify: 'TIMEOUT: No changes in 30s. Retry or use wechat_read() for on-demand fetch.'
• Use JSON Schema allOf + if/then to express conditional parameter requirements, or document them in a 'Parameter Dependencies' section of the tool description. Example: 'If kind=file, path is required. If kind=sticker and collection=search, query is required.'
• Add pagination to wechat_read: accept offset/limit, return a total count or next_signature, and document: 'Use next_signature with after= in a follow-up call to fetch the next page of messages.'
• Document idempotency guarantees: For wechat_send, clarify whether sending the same text to the same chat twice produces one or two messages. If deduplication is based on a timestamp window, document it. If not idempotent, explain the risk to agents (duplicate messages).