sharaf/speech-recognition-architect
Use when the user wants to design, size, audit, or choose a self-hosted speech recognition or streaming ASR stack, including Whisper, Parakeet, Canary, Riva, NIM, Triton ASR, faster-whisper, sherpa-onnx, voice-agent transcription, Romanian or Moldovan ASR, contact-center transcription, GPU sizing, latency budgets, multilingual routing, VAD, diarization, or production evaluation.
100
100%
Does it follow best practices?
Impact
100%
2.00xAverage score across 3 eval scenarios
Passed
No known issues
Quality
Discovery
100%Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.
This is a strong skill description that clearly defines a specific niche (self-hosted speech recognition/ASR) with comprehensive trigger terms covering technologies, tasks, and use cases. It explicitly uses the 'Use when...' pattern and lists concrete actions (design, size, audit, choose). The only minor weakness is that it reads somewhat like a keyword dump, but the keywords are all highly relevant and natural for the domain.
| Dimension | Reasoning | Score |
|---|---|---|
Specificity | The description lists multiple specific concrete actions: 'design, size, audit, or choose a self-hosted speech recognition or streaming ASR stack' and names specific technologies (Whisper, Parakeet, Canary, Riva, NIM, Triton ASR, faster-whisper, sherpa-onnx) and tasks (GPU sizing, latency budgets, multilingual routing, VAD, diarization, production evaluation). | 3 / 3 |
Completeness | The description explicitly starts with 'Use when the user wants to...' which clearly answers both 'what' (design/size/audit/choose self-hosted ASR stacks with specific technologies and tasks) and 'when' (when the user mentions any of the listed topics or technologies). | 3 / 3 |
Trigger Term Quality | Excellent coverage of natural terms users would say: 'speech recognition', 'ASR', 'Whisper', 'transcription', 'voice-agent', 'contact-center', 'GPU sizing', 'diarization', 'VAD', plus specific product names like 'Riva', 'Triton', 'faster-whisper', 'sherpa-onnx'. These are terms practitioners would naturally use. | 3 / 3 |
Distinctiveness Conflict Risk | Highly distinctive niche focused on self-hosted speech recognition/ASR stacks with very specific technology names and domain terms (Romanian/Moldovan ASR, contact-center transcription, GPU sizing for ASR). Unlikely to conflict with other skills unless there's another ASR-specific skill. | 3 / 3 |
Total | 12 / 12 Passed |
Implementation
100%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This is an exceptionally well-crafted skill that provides dense, expert-level architectural guidance for self-hosted ASR systems. Its strengths are the specific decision anchors with concrete model names, rejection rationales, and configuration details that Claude wouldn't inherently know, combined with a clear workflow and precise output contract. The skill respects Claude's intelligence while adding substantial domain-specific value around Romanian ASR, GPU sizing math, and production evaluation criteria.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The skill is dense and efficient. Every section earns its place—decision anchors, sizing formulas, evaluation criteria, and output contract are all domain-specific knowledge Claude wouldn't have. No unnecessary explanations of what ASR is or how GPUs work. | 3 / 3 |
Actionability | Highly actionable despite being a design-oriented (not code-oriented) skill. Decision anchors provide specific model names, rejection reasons, concrete configuration parameters (e.g., `model_transaction_policy { decoupled: true }`), exact sizing formulas, and a precise output contract template. The guidance is concrete and copy-paste ready for architecture documents. | 3 / 3 |
Workflow Clarity | The 6-step workflow is clearly sequenced from requirements gathering through blueprint output. Validation is embedded via evaluation anchors (WER benchmarks, silence hallucination probes, RTFx), the 'Done When' section acts as a completion checklist, and the guardrails provide clear boundaries. The eval gates serve as feedback loops for verifying design decisions. | 3 / 3 |
Progressive Disclosure | For a standalone skill with no bundle files, the content is well-organized into clearly labeled sections (Decision Anchors, Sizing Anchors, Evaluation Anchors, Blueprint Output Contract, Guardrails, Done When). The table format for decision anchors is efficient. No monolithic walls of text, and the content length is appropriate for the complexity without needing external files. | 3 / 3 |
Total | 12 / 12 Passed |
Validation
100%Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.
Validation — 11 / 11 Passed
Validation for skill structure
No warnings or errors.
Reviewed
Table of Contents