Gateway tools have clear verb-noun naming and reasonable descriptions (avg 120 chars), but parameter documentation is inconsistent. Skills tools lack descriptions for the 'command' parameter in skills__marketplace_install. No output schemas are documented in the source code. Error handling guidance is absent, tools do not indicate what to do on failure or whether errors are retryable. Tool composition is reasonable (each does one thing), but chaining support is weak, responses do not document what IDs/references downstream tools need.

• Document output schemas for all 7 tools. Specify field names, types (string, number, array, object), and what each field represents. Example: gateway__list_servers should return {servers: [{key: string, status: 'connected'|'disconnected', tool_count: number, ...}], total: number}.
• Add descriptions to all parameters. For skills__marketplace_install, clarify: 'The install command from skills.sh marketplace (e.g., "npx skills add https://github.com/owner/repo --skill skill-name"). Must include the --skill flag.'
• Add error handling guidance to write tools. E.g., skills__marketplace_install: 'On failure, returns error with reason (network, invalid command, repo not found). Errors are not retryable if the command syntax is invalid; retry if network timeout.'
• Explicitly mark write operations in descriptions. Prefix skills__sync and skills__marketplace_install descriptions with 'Modifies state:' to signal to LLMs that these are irreversible.
• Document response chaining. Specify what IDs/references each tool returns so downstream tools can use them. E.g., 'skills__list returns skill_id for each skill; pass skill_id to skills__get.'

Inferred effective spec: <=2025-11-25.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained