Single tool 'transcribe_audio' has a clear action verb and basic schema, but lacks depth in descriptions and parameter documentation. Tool description is 28 chars (below 34 p10 baseline), parameter descriptions are minimal, and no output schema is documented. No error handling guidance, no recovery paths, and no indication of what fields the response contains. The schema is present but sparse, file type is correct, but diarization and language lack detail on constraints or use cases.

• Expand tool description to 100 - 150 chars: 'Transcribe an audio file using Gladia API. Supports speaker diarization and language detection. Returns transcription text, word-level timing, confidence scores, and speaker labels. Requires GLADIA_API_KEY environment variable.'
• Document the output schema inline or in a separate schema definition: 'Returns {status: string, result: {language, text, words: [{word, start, end, confidence, speaker}], duration, processing_time}, result_url, error}'
• Add parameter descriptions: language → 'ISO 639-1 language code (e.g. en, fr, es). If null, language is auto-detected.'; diarization → 'If true, identifies and labels different speakers in the audio.'
• Add error handling guidance: 'If transcription fails, check that GLADIA_API_KEY is valid and the audio file is in a supported format (MP3, WAV, OGG, FLAC). Retry after 5 seconds if the service is temporarily unavailable.'
• Add tool annotations to the schema: mark as readOnlyHint=true and idempotentHint=true so LLM knows it is safe to retry and does not modify state.

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

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