Three tools with mixed quality. Agent and AskUserQuestion have solid descriptions (100+ chars) and clear purposes, but lack visible input schema validation and error handling guidance. jira_search has good description with practical constraints (unbounded query warning, filter requirements) and documented output fields, but schema completeness unclear. No tool shows parameter-level descriptions in the visible code. Agent tool description mentions recursion depth-limiting but no error recovery guidance. Overall: descriptions present and actionable, but schemas and error handling need strengthening.

• Add parameter-level descriptions to all three tools' input schemas. For Agent: describe what 'subagent_type' values are valid (reference ListSubagentTypes), explain 'model' override behavior. For AskUserQuestion: document timeout behavior and default. For jira_search: explain max_results default and cap in the parameter description, not just the tool description.
• Document error responses for Agent: what error is returned if recursion depth is exceeded? Is it retryable? Add guidance like 'If recursion limit hit, break the task into smaller sub-prompts or use a different subagent_type.'
• Add output schema documentation for all tools. Specify what fields Agent returns (is it a string, object, or structured result?). Document jira_search return structure (confirm it returns key, summary, status, type, priority, assignee, updated as stated).
• For jira_search, enhance error handling: document what happens when a query is rejected as unbounded. Return a specific error message like 'Query rejected: must include a filter (project, assignee, or updated >= -30d). Your query: {jql}' so the LLM can self-correct.
• Add idempotency and side-effect guidance: Agent description should clarify that spawning a sub-agent is a write operation with token/time cost and is not idempotent. AskUserQuestion should note it blocks until user responds (timeout or response). jira_search is read-only and safe to retry.