Add explicit instruction to use 'gh pr comment' command to post
reviews to the PR. Claude was performing reviews but not posting
them without this explicit direction.
- Create .github/slash-commands/ with link-review, model-check, notebook-review
- Update GitHub Actions to use slash commands instead of inline prompts
- Add symlinks in .claude/commands/ for local development
- Document slash commands in CONTRIBUTING.md
- Use claude-code-action@v1 instead of beta
This allows developers to run the same CI validations locally using
Claude Code slash commands before pushing changes.
Major simplification of CI/CD:
- Remove complex Python model validation scripts (400+ lines)
- Let Claude handle model validation intelligently via GitHub Actions
- Claude fetches latest models from docs.anthropic.com/en/docs/about-claude/models/overview.md
- Add comprehensive notebook validation script for local testing
- Interactive dashboard with progress tracking
- Auto-fix for deprecated models
- GitHub issue export format
- Idempotent with state persistence
- Simplify CI to use single Python version (3.11)
- Update workflows to use Claude for all intelligent validation
Benefits:
- No more hardcoded model lists to maintain
- Claude understands context (e.g., educational examples)
- 50% faster CI (removed matrix strategy)
- Single source of truth for models (docs site)
- Brittle check that will break with new model versions
- Claude already provides intelligent model recommendations
- Not all notebooks should use Haiku (some need more capability)
- Use Python 3.11 only (no version-specific code in notebooks)
- Reduces CI runtime by 50%
- Reduces API costs by 50% for notebook execution
- Simplifies PR checks (one instead of two identical)
- Add skip_code_blocks=true (important for notebooks with example code)
- Add require_https=false for development flexibility
- Simplify accept codes to just 403 and 429 (like docs)
- Add www.claude.ai to exclusions
- Exclude .github/ and scripts/ paths from checking
- Better comments explaining each setting
- Remove brittle hardcoded API key checks from validate_notebooks.py
- Enhance Claude review to check for any secrets (not just Anthropic)
- Claude understands context (e.g., educational 'bad examples' are OK)
- Check for clear introduction explaining the notebook's purpose
- Validate configuration instructions are present
- Ensure connecting explanations between cells for better flow
- Add claude-notebook-review.yml for intelligent code review
- Add claude-link-review.yml for link quality checks
- Update notebook-quality.yml to properly capture test outputs
- Use anthropics/claude-code-action@beta like docs repo
Notebook outputs are educational content in cookbook repositories.
They show users what to expect when running the code.
- Remove nbstripout from all dependencies and configurations
- Remove nbstripout check from CI workflow
- Update documentation to explain outputs are intentional
- Make validation scripts non-blocking for POC
- Fix lychee configuration conflict
The CI now validates notebooks without removing demonstration outputs.
- Make ruff checks non-blocking (|| true) for notebooks
- Make model validation report issues but not fail
- This allows the POC to demonstrate issue detection without blocking
The CI now shows what issues exist without preventing the PR from
being merged, which is appropriate for a proof of concept.
Remove nbqa in favor of ruff's native Jupyter support (v0.6.0+).
Replace papermill with nbconvert due to uv dependency resolution issues.
Also remove S105/S106 ignores to enforce better security practices.
- Update pyproject.toml to use ruff v0.12.12 with native notebook support
- Replace papermill with nbconvert for notebook execution
- Remove nbqa from all dependencies and pre-commit hooks
- Update GitHub Actions workflows to use ruff directly
- Remove hardcoded password ignores for better security
- Update documentation to reflect simplified setup
- Add dummy package structure for hatchling build system
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Configure lychee for notebook link validation
- Set up GitHub workflow for PR and scheduled checks
- Exclude API endpoints and localhost from checks
- Add PR comment integration for broken links
- Add nbstripout to clean notebook outputs
- Configure nbqa with ruff for notebook linting
- Add ruff for Python file formatting
- Add custom hooks for model and notebook validation
- Add notebook-quality.yml with papermill execution testing
- Add claude-model-check.yml for model validation
- Add security-scan.yml for secret detection
- Implement tiered testing (full for maintainers, mock for external)
- Add allowed_models.py that fetches latest models from Anthropic docs
- Implement 24-hour cache to avoid excessive requests
- Fall back to hardcoded list if fetch fails
- Add check_models.py to validate model usage in notebooks
- Add validate_notebooks.py for security and structure checks
- Update .gitignore for cache files
Update timestamp generation to snap to bucket boundaries to ensure
consistent data retrieval, as partial dates may not return data.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Comprehensive guide to programmatically accessing Claude API usage and cost data for custom dashboards, cost monitoring, and usage analysis.
Features:
- Basic usage and cost tracking
- Time granularity and filtering options
- Grouping and breakdowns
- Priority Tier analysis
- Pagination for large datasets
- Error handling best practices
- Practical alerting examples
