Summarize CLI

Use this skill to run summarize effectively in terminal workflows. Focus on practical usage and command execution, not broad project background.

Workflow

1. Confirm input type and desired output format.
2. Run a minimal real command first (summarize "<input>").
3. Select a model/provider and ensure matching API credentials.
4. Add flags incrementally only after a baseline run succeeds.
5. If it fails, re-run with --verbose or --json before changing multiple variables.

Quick Patterns

• Summarize a URL: summarize "https://example.com"
• Summarize a local file: summarize "/path/to/file.pdf"
• Summarize piped input: cat notes.md | summarize -
• Summarize YouTube/media URL: summarize "https://youtube.com/watch?v=..."
• Force model/provider: summarize "https://example.com" --model openai/gpt-5-mini
• Control length: summarize "https://example.com" --length long
• Return diagnostics JSON: summarize "https://example.com" --json
• Extract content without summarizing: summarize "https://example.com" --extract --format md

Configuration Rules

• Prefer explicit --model on task-critical runs.
• Use ~/.summarize/config.json for stable defaults.
• Remember precedence for model selection:
  1. --model
  2. SUMMARIZE_MODEL
  3. ~/.summarize/config.json
  4. built-in default (auto)

Troubleshooting Flow

1. Re-run with --verbose.
2. Confirm API key matches the chosen provider/model.
3. Switch to a known-good model (for example openai/gpt-5-mini or google/gemini-3-flash-preview).
4. For media or YouTube failures, verify external dependencies (yt-dlp, ffmpeg, optionally tesseract).
5. Use --json to inspect extraction and metric fields.

Scope

• Prioritize how to run commands and fix concrete failures.
• Do not include full repository-level documentation unless explicitly requested.
• Use references/cli.md only for deeper flag/provider details.

References

Read references/cli.md for install commands, provider keys, advanced flags, and error-specific guidance.