Six tools with clear verb-noun naming (watch_video, get_transcript, search_transcript, get_frames, get_frame_at, list_videos). All have descriptions (194 - 250 chars, within baseline 34 - 392). Input schemas present with type declarations and parameter descriptions. However, output schemas are not documented in the visible code, LLMs cannot predict return structure. Error handling is minimal; no recovery guidance or actionable error messages visible. watch_video supports long-polling (idempotent), but no confirmation pattern for destructive operations. Parameter constraints (model enum, device enum, language code) are described but not formally declared as JSON Schema enums.

• Document output schemas for all tools. Example: watch_video returns {status: 'RUNNING'|'DONE', job_id, progress_percent, transcript_path, frames_dir} or {status: 'DONE', transcript: string, keyframes: [image], report_url}.
• Convert parameter constraints to JSON Schema enums. watch_video.model should be enum: ['auto', 'tiny', 'base', 'small', 'medium', 'large-v3', 'large-v3-turbo']; device should be enum: ['auto', 'cuda', 'cpu'].
• Add error handling guidance. E.g., 'If watch_video returns an error about unsupported format, try a different video URL or check that the file is a valid MP4/WebM. For GPU out-of-memory, set device=cpu or model=small.'
• Enhance list_videos description to explain return structure (e.g., 'Returns array of {video_id, source_url, transcript_length, frame_count, processed_at}') and add optional filter/limit parameters.
• Add pagination to list_videos: accept limit (1 - 100, default 20) and offset or cursor parameters. Return total_count and next_cursor in response.

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

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry