davidtoby Model CLI and Review Workflows
Use this umbrella when the main task involves invoking another model/provider tool or auditing model usage.
Core workflow
- Identify the model/provider CLI and whether the task is generation, review, transcription/image generation, retrieval, or cost analysis.
- Check authentication and environment variables before long runs.
- Bundle prompts with only the necessary files/context; avoid leaking secrets.
- Capture command, model, inputs, and output path/summary.
- Verify outputs exist and are in the expected format before reporting success.
Labeled playbooks
Gemini CLI
Use for one-shot Q&A, summarization, and generation when a Gemini model is specifically desired or useful as an independent perspective.
Oracle / second-model review
Bundle focused prompts plus relevant files and ask for debugging/refactor/design review. Treat the answer as advisory; verify claims locally.
Model usage/cost
Use local usage/cost CLIs to summarize current model, model breakdowns, and recent spend. Never infer costs without reading usage data.
qmd/local retrieval
Index/search local docs when a task needs retrieval across a corpus before model synthesis.
OpenAI media endpoints
Use the dedicated media umbrella for broader media workflows; keep provider-specific API quirks here as references.
Reference files
Exact command forms and provider quirks are stored under references/from-*.md.