The ai-tools subject provides CLI tools for using AI to edit and refine documentation based on prompt-driven guidelines. It integrates with GitHub Models API to apply automated improvements to content files.

This subject is responsible for:

• AI-powered content refinement (versioning, intro, etc.)
• Prompt-driven content editing with LLMs
• Integration with GitHub Models API
• Copilot Spaces export and conversion to prompts
• Automated content quality improvements
• Extensible prompt system for different refinement types

Current refinements: versioning, intro. Future: scannability, readability, style, technical accuracy.

• scripts/ai-tools.ts - Main CLI tool for running AI refinements
• lib/call-models-api.ts - Client for GitHub Models API inference
• lib/prompt-utils.ts - Loads prompts and executes refinements
• prompts/*.md - Prompt templates for different refinement types

Add GitHub token with Models scopes to .env:

GITHUB_TOKEN=ghp_your_token_here

# Direct command
tsx src/ai-tools/scripts/ai-tools.ts --refine versioning --files content/path/to/file.md

# Via npm script
npm run ai-tools -- --refine versioning --files content/path/to/file.md

• --files, -f: One or more content file paths (required)
• --refine, -r: Refinement type(s) - versioning, intro (default: versioning)
• --write, -w: Write changes to files (default: false, shows diff only)
• --verbose, -v: Verbose output for debugging
• --space, -s: Use Copilot Space as prompt
• --exportSpace: Export Copilot Space to prompt file

Refine versioning in a file:

npm run ai-tools -- --files content/copilot/tutorials/coding-agent/get-the-best-results.md --refine versioning

Refine intro:

npm run ai-tools -- --files content/pull-requests/collaborating-with-pull-requests/working-with-forks/fork-a-repo.md --refine intro

Multiple files:

npm run ai-tools -- --files file1.md file2.md file3.md --refine versioning

Write changes (not just preview):

npm run ai-tools -- --files content/path/to/file.md --refine versioning --write

• Content markdown files with frontmatter
• Prompt templates in prompts/ directory
• GitHub Models API for inference

• GitHub Models API - Requires GITHUB_TOKEN with Models scopes
• commander - CLI argument parsing
• dotenv - Environment variable loading
• Copilot Spaces (optional) - Can export/import prompts

• Refined markdown content (preview or written to files)
• Diffs showing proposed changes
• Merged frontmatter updates

• src/content-render - Content parsing and rendering
• Content files in content/ - Target of refinements

• GitHub Models API docs
• Copilot Spaces for prompt management

• Team: Docs Content (for use and development)
• Note: These tools are for the docs-content team. We welcome them to use Copilot to support and develop these tools, but docs-engineering is largely hands off.

Current prompts:

• versioning - Refines version-related content
• intro - Improves article introductions

Each prompt defines:

• Instructions for the LLM
• Expected output format
• Quality criteria

1. Create prompt file in prompts/ (e.g., readability.md).
2. Write prompt instructions and examples.
3. Test with Models UI first.
4. Use --refine readability to apply.

Prompt template in prompts/prompt-template.yml.

Export Space to prompt:

npm run ai-tools -- --exportSpace space-id --output prompts/my-prompt.md

Use Space as prompt:

npm run ai-tools -- --space space-id --files content/path/to/file.md

• Requires GitHub token with Models scopes
• API rate limits apply
• Quality depends on prompt engineering
• Currently manual execution (not in CI)
• No automated testing/evals yet

Prompt engineering:

• Test prompts in GitHub Models UI first
• Include clear examples in prompts
• Define expected output format
• Iterate on prompts based on results

File selection:

• Start with single files to test
• Use glob patterns for batch processing
• Preview changes before writing

Quality assurance:

• Always review AI suggestions
• Don't blindly accept all changes
• Consider subject matter expertise needed
• Test refined content for correctness

Missing token error: Add GITHUB_TOKEN to .env with Models scopes.

API errors:

• Check token permissions
• Verify rate limits
• Check Models API status

Poor refinement quality:

• Refine the prompt template
• Add more examples
• Test in Models UI first
• Consider different model/parameters