openai-whisper-api ยท diff

v1.0.0 to v1.1.0

6 added, 6 removed. Audit A to A.

---
name: openai-whisper-api
- description: "Transcribe audio via OpenAI Audio Transcriptions API (Whisper)."
- version: "1.0.0"
+ description: "Transcribe audio via OpenAI Whisper or Atlas Cloud speech-to-text APIs."
+ version: "1.1.0"
metadata:
author: CoWork OS Contributors <info@coworkosapp.com>
---
# Openai-whisper-api
## Purpose
- Transcribe audio via OpenAI Audio Transcriptions API (Whisper).
+ Transcribe audio via OpenAI Whisper or Atlas Cloud speech-to-text APIs.
## Routing
- - Use when: Use when the user asks to transcribe audio via OpenAI Audio Transcriptions API Whisper.
+ - Use when: Use when the user asks to transcribe audio through OpenAI Whisper or Atlas Cloud speech-to-text.
- Do not use when: Do not use when the request is asking for planning documents, high-level strategy, or non-executable discussion; use the relevant planning or design workflow instead.
- Outputs: Outcome from Openai-whisper-api: task-specific result plus concrete action notes.
- Success criteria: Returns concrete actions and decisions matching the requested task, with no fabricated tool-side behavior.
## Trigger Examples
### Positive
- Use the openai-whisper-api skill for this request.
- Help me with openai-whisper-api.
- - Use when the user asks to transcribe audio via OpenAI Audio Transcriptions API Whisper.
+ - Use when the user asks to transcribe audio through OpenAI Whisper or Atlas Cloud speech-to-text.
- Openai-whisper-api: provide an actionable result.
### Negative
- Do not use when the request is asking for planning documents, high-level strategy, or non-executable discussion; use the relevant planning or design workflow instead.
- Do not use openai-whisper-api for unrelated requests.
- This request is outside openai-whisper-api scope.
- This is conceptual discussion only; no tool workflow is needed.
## Runtime Prompt
- - Current runtime prompt length: 796 characters.
+ - Current runtime prompt length: 1245 characters.
- Runtime prompt is defined directly in `../openai-whisper-api.json`.