Two tools with well-structured schemas and detailed descriptions. authloop_to_human has comprehensive parameter documentation with nested context object and enum constraints (blocker_type). authloop_status has minimal input but clear blocking semantics. Both descriptions exceed 100 chars and explain WHEN/WHY to use. However, output schemas are not explicitly documented in the source code provided, only inferred from descriptions. Tool names use verb_noun pattern (authloop_to_human, authloop_status) but lack the standard action verb prefix (send_, wait_) that would improve LLM parsing. No error handling guidance visible in descriptions.

• Document output schemas explicitly. For authloop_to_human, define: {session_url: string, expires_at?: ISO8601}. For authloop_status, define: {status: 'resolved'|'cancelled'|'timeout'|'error', error_message?: string, resolved_at?: ISO8601}.
• Rename tools to standard verb_noun pattern: authloop_to_human → send_to_human_solver or delegate_to_human; authloop_status → wait_for_human_response or poll_session_status.
• Add error handling section to authloop_status description: 'If timeout occurs, the session is cancelled. Call authloop_to_human again to create a new session. Retries are safe (idempotent).'
• Move procedural instructions out of tool descriptions. Instead, document in a separate README or agent prompt. Tool descriptions should state WHAT the tool does, not HOW the agent should orchestrate multiple tools.
• For context parameter, add guidance: 'Recommended: always include url and blocker_type to help the human understand the challenge. hint is optional but improves UX for OTP/security questions.'