- Update all LLM model files to use getSecretSync with AI_KEYS constants
- Migrate OpenAI GPT models (gpt-5, gpt-5-mini, gpt-5-nano) to use secrets system
- Migrate Anthropic models (sonnet-4, haiku-3-5) to use secrets for both Anthropic API and Vertex AI credentials
- Update provider files (openai.ts, anthropic.ts, vertex.ts) to use getSecretSync
- Migrate evaluation workflow to use BRAINTRUST_KEY from secrets
- Implement proper error handling for missing credentials with try-catch blocks
- Test files will be migrated in a separate update
All non-test source files now use centralized secrets management instead of direct process.env access.
- Introduced GitHub routes for handling installation callbacks and token management.
- Implemented middleware for validating GitHub webhook requests.
- Added tests for new handlers and middleware to ensure functionality.
- Enhanced sandbox utilities to support GitHub token usage for authenticated operations.
- Introduced integration tests for various file tools including bash execution, file creation, deletion, editing, and listing.
- Enhanced the check-off todo list tool to support multiple items and improved error handling for non-existent tasks.
- Updated test cases to cover edge scenarios and ensure robust validation across all tools.
- Increased timeout for the docs agent workflow to accommodate longer-running tests.
- Added section in CLAUDE.md for direct database access during integration testing.
- Updated `maxSteps` in `docs-agent` to allow for more complex tasks.
- Improved validation in `docs-agent-context` for sandbox instances.
- Enhanced `create-docs-todos` step to handle todos more effectively.
- Introduced comprehensive integration tests for the docs agent workflow, covering various scenarios and edge cases.
- Added test helpers for creating mock dbt projects and managing sandboxes.
- Implemented error handling and logging improvements in the workflow execution process.
- Add read_files tool in packages/ai/src/tools/file-tools
- Support both absolute and relative file paths
- Implement 1000-line truncation with indication
- Handle errors gracefully with discriminated union results
- Include comprehensive unit tests with 10/10 passing
- Export tool from main index
- Update sandbox index to export runTypescript and createSandbox
- Follow established patterns from execute-sql.ts
Co-Authored-By: Dallin Bentley <dallinbentley98@gmail.com>
dal
Nate Kelley
Devin AI