18 tools with schemas present but significant quality gaps. All tools have descriptions (positive), but most are under 50 chars and lack actionable context. Parameter descriptions are minimal (e.g., '可选的文件夹路径' = 'optional folder path'). No output schemas documented. No error handling guidance. No tool annotations (readOnlyHint/destructiveHint). Naming is verb-first (getJobList, createJob) but descriptions are in Chinese, limiting LLM comprehension in English-first agents. Risk classifications visible in metadata but not in tool descriptions themselves. Missing idempotency guidance for destructive ops (deleteJob, killJob).

• Translate all descriptions and parameter docs to English. Use 50 - 150 char descriptions that state WHAT the tool does, WHEN to use it, and what it returns. E.g., 'List all Jenkins jobs in a folder. Returns job names, status, and last build time. Call this first to discover available jobs.'
• Document output schemas for all tools. E.g., getJobList returns [{name, status, lastBuild, url}]. getBuildLog returns {log: string, hasMore: boolean, nextStart: number}.
• Add tool annotations to schema: readOnlyHint=true for get_* and list_* tools; destructiveHint=true for delete_* and kill_*; idempotentHint=true for enable/disable (safe to retry).
• For destructive tools (deleteJob, killJob), add a confirmation parameter or implement a two-step pattern: first call returns a confirmation_token, second call requires that token. Document this in the tool description.
• Add parameter constraints: buildNumber must be integer ≥1; start must be integer ≥0; changeSetFlag must be 'true'|'false'. Use enums or regex patterns in schema, and repeat constraints in descriptions.