- Removed redundant definition of CLI base timeout in `cli-provider.ts` and added a detailed comment explaining its purpose.
- Updated `codex-provider.ts` to use the imported `DEFAULT_TIMEOUT_MS` directly instead of an alias.
- Enhanced unit tests to ensure fallback behavior for invalid reasoning effort values in timeout calculations.
- Added `calculateReasoningTimeout` function to dynamically adjust timeouts based on reasoning effort levels.
- Updated CLI and Codex providers to utilize the new timeout calculation, addressing potential timeouts for high reasoning efforts.
- Enhanced unit tests to validate timeout behavior for various reasoning efforts, ensuring correct timeout values are applied.
CLI-based providers (OpenCode, etc.) only accept a single prompt via
stdin/args and don't support separate system/user message channels like
Claude SDK. When systemPrompt is passed to these providers, it was
silently dropped, causing:
- BacklogPlan JSON parsing failures with OpenCode/GPT-5.2 (missing
"output ONLY JSON" formatting instruction)
- Loss of critical formatting/schema instructions for structured outputs
This fix adds embedSystemPromptIntoPrompt() method to CliProvider base
class that:
- Prepends systemPrompt to the user prompt before CLI execution
- Handles both string and array prompts (vision support)
- Handles both string systemPrompt and SystemPromptPreset objects
- Uses standard \n\n---\n\n separator (consistent with codebase)
- Sets systemPrompt to undefined to prevent double-injection
Benefits OpencodeProvider immediately (uses base executeQuery).
CursorProvider still uses manual workarounds (overrides executeQuery).
Fixes the immediate BacklogPlan + OpenCode bug while maintaining
backward compatibility with existing Cursor workarounds.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Skip learning extraction when ANTHROPIC_API_KEY is not available
- Add reasoningEffort parameter to simpleQuery for Codex model configuration
- Add stdinData support to spawnProcess for CLI stdin input
- Update UI API types for model override with reasoning support
When an OpenAI API key is stored in settings or environment, use SDK mode
instead of CLI mode. This bypasses the MCP transport layer which was
failing with 'TokenRefreshFailed' errors due to OAuth token issues.
The SDK uses the API key directly via @openai/codex-sdk, avoiding the
OAuth token refresh mechanism that was causing mid-execution failures.
- Introduced a new Simple Query Service to streamline basic AI queries, allowing for structured JSON outputs.
- Updated existing routes to utilize the new service, replacing direct SDK calls with a unified interface for querying.
- Enhanced provider handling in various routes, including generate-spec, generate-features-from-spec, and validate-issue, to support both Claude and Cursor models seamlessly.
- Added structured output support for improved response handling and error management across the application.
- Update isOpencodeModel() to detect dynamic models with provider/model format
(e.g., github-copilot/gpt-4o, google/gemini-2.5-pro, zai-coding-plan/glm-4.7)
- Update resolveModelString() to recognize and pass through OpenCode models
- Update enhance route to route OpenCode models to OpenCode provider
- Fix OpenCode CLI command format: use --format json (not stream-json)
- Remove unsupported -q and - flags from CLI arguments
- Update normalizeEvent() to handle actual OpenCode JSON event format
- Add dynamic model configuration UI with provider grouping
- Cache providers and models in app store for snappier navigation
- Show authenticated providers in OpenCode CLI status card
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Fix TS2322 error where finishEvent.part?.result (typed as {}) was being
assigned to result.result (typed as string).
Solution: Safely handle arbitrary result payloads by:
1. Reading raw value as unknown from Record<string, unknown>
2. Checking if it's a string, otherwise JSON.stringify()
This ensures type safety while supporting both string and object results
from the OpenCode CLI.
Fix all 8 remaining test failures:
1. Update executeQuery integration tests to use new OpenCode event format:
- text events use type='text' with part.text
- tool_call events use type='tool_call' with part containing call_id, name, args
- tool_result events use type='tool_result' with part
- step_finish events use type='step_finish' with part
- Use sessionID field instead of session_id
2. Fix step_finish event handling:
- Include result field in successful completion response
- Check for reason === 'error' to detect failed steps
- Provide default error message when error field is missing
3. Update model test expectations:
- Model 'opencode/big-pickle' stays as-is (not stripped to 'big-pickle')
- PROVIDER_PREFIXES only strips 'opencode-' prefix, not 'opencode/'
All 84 tests now pass successfully!
This commit integrates OpenCode as a new AI provider and updates all provider
icons with their official brand colors for better visual recognition.
**OpenCode Provider Integration:**
- Add OpencodeProvider class with CLI-based execution
- Support for OpenCode native models (opencode/) and Bedrock models
- Proper event normalization for OpenCode streaming format
- Correct CLI arguments: --format json (not stream-json)
- Event structure: type, part.text, sessionID fields
**Provider Icons:**
- Add official OpenCode icon (white square frame from opencode.ai)
- Add DeepSeek icon (blue whale #4D6BFE)
- Add Qwen icon (purple gradient #6336E7 → #6F69F7)
- Add Amazon Nova icon (AWS orange #FF9900)
- Add Mistral icon (rainbow gradient gold→red)
- Add Meta icon (blue #1877F2)
- Update existing icons with brand colors:
* Claude: #d97757 (terra cotta)
* OpenAI/Codex: #74aa9c (teal-green)
* Cursor: #5E9EFF (bright blue)
**Settings UI Updates:**
- Update settings navigation to show OpenCode icon
- Update model configuration to use provider-specific icons
- Differentiate between OpenCode free models and Bedrock-hosted models
- All AI models now display their official brand logos
**Model Resolution:**
- Add isOpencodeModel() function to detect OpenCode models
- Support patterns: opencode/, opencode-*, amazon-bedrock/*
- Update getProviderFromModel to recognize opencode provider
Note: Some unit tests in opencode-provider.test.ts need updating to match
the new event structure and CLI argument format.
- Add error logging to CodexProvider auth check instead of silent failure
- Fix cachedAt timestamp to return actual cache time instead of request time
- Replace misleading hardcoded rate limit values (100) with sentinel value (-1)
- Fix unused parameter warning in codex routes
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Bumped version numbers for @automaker/server and @automaker/ui to 0.9.0 in package-lock.json.
- Introduced CodexAppServerService and CodexModelCacheService to manage communication with the Codex CLI's app-server and cache model data.
- Updated CodexUsageService to utilize app-server for fetching usage data.
- Enhanced Codex routes to support fetching available models and integrated model caching.
- Improved UI components to dynamically load and display Codex models, including error handling and loading states.
- Added new API methods for fetching Codex models and integrated them into the app store for state management.
These changes improve the overall functionality and user experience of the Codex integration, ensuring efficient model management and data retrieval.
- Combined CLI disconnection markers with OpenCode support
- Added OpenCode auth/deauth routes and API methods
- Resolved merge conflicts between feature branch and upstream v0.9.0rc
- Fix Claude, Codex, and Cursor auth handlers to check if CLI is already authenticated
- Use same detection logic as each provider's internal checkAuth/codexAuthIndicators()
- For Codex: Check for API keys and auth files before requiring manual login
- For Cursor: Check for env var and credentials files before requiring manual auth
- For Claude: Check for cached auth tokens, settings, and credentials files
- If CLI is already authenticated: Just reconnect by removing disconnected marker
- If CLI needs auth: Tell user to manually run login command
- This prevents timeout errors when login commands can't run in non-interactive mode
- Added OpenCode authentication status check to the OpencodeProvider class.
- Introduced OpenCodeAuthStatus interface to manage authentication states.
- Updated detectInstallation method to include authentication status in the response.
- Created ProvidersSetupStep component to consolidate provider setup UI, including Claude, Cursor, Codex, and OpenCode.
- Refactored setup view to streamline navigation and improve user experience.
- Enhanced OpenCode CLI integration with updated installation paths and authentication checks.
This commit enhances the setup process by allowing users to configure and authenticate multiple AI providers, improving overall functionality and user experience.
- Implemented OpenCode CLI installation and authentication status check.
- Added new route for OpenCode status in setup routes.
- Updated HttpApiClient to include method for fetching OpenCode status.
- Enhanced system paths to include OpenCode's default installation directories.
This commit introduces functionality to check the installation and authentication status of the OpenCode CLI, improving integration with the overall system.
Resolved conflict in agent-service.ts by keeping both:
- agents parameter for custom subagents (from our branch)
- thinkingLevel and reasoningEffort parameters (from v0.9.0rc)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Moves prefix stripping from individual providers to AgentService/IdeationService
and adds validation to ensure providers receive bare model IDs. This prevents
bugs like the Codex CLI receiving "codex-gpt-5.1-codex-max" instead of the
expected "gpt-5.1-codex-max".
- Add validateBareModelId() helper with fail-fast validation
- Add originalModel field to ExecuteOptions for logging
- Update all providers to validate model has no prefix
- Centralize prefix stripping in service layer
- Remove redundant prefix stripping from individual providers
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Fix isCursorModel to exclude Codex-specific models (gpt-5.1-codex-*, gpt-5.2-codex-*)
- These models should route to Codex provider, not Cursor provider
- Add CODEX_YOLO_FLAG constant for --dangerously-bypass-approvals-and-sandbox
- Always use YOLO flag in codex-provider for full permissions
- Simplify codex CLI args to minimal set with YOLO flag
- Update tests to reflect new behavior with YOLO flag
This fixes the bug where selecting a Codex model (e.g., gpt-5.1-codex-max)
was incorrectly spawning cursor-agent instead of codex exec.
The root cause was:
1. Cursor provider had higher priority (10) than Codex (5)
2. isCursorModel() returned true for Codex models in CURSOR_MODEL_MAP
3. Models like gpt-5.1-codex-max routed to Cursor instead of Codex
The fix:
1. isCursorModel now excludes Codex-specific model IDs
2. Codex always uses --dangerously-bypass-approvals-and-sandbox flag
- Added a new utility for checking Codex CLI authentication status using the 'codex login status' command.
- Integrated the authentication check into the CodexProvider's installation detection and authentication methods.
- Updated Codex CLI status display in the UI to reflect authentication status and method.
- Enhanced error handling and logging for better debugging during authentication checks.
- Refactored related components to ensure consistent handling of authentication across the application.
- Changed the event type from 'thread.completed' to 'turn.completed' in the CODEX_EVENT_TYPES constant and its usage within the CodexProvider class.
- This update aligns the event handling with the intended functionality, ensuring correct event processing.
- Removed unnecessary JSON.stringify conversion for string values in formatConfigValue function, streamlining the value formatting process.
- This change enhances code clarity and reduces complexity in the configuration handling of the CodexProvider.
- Added approval policy and web search features to the CodexProvider's argument construction, improving flexibility in command execution.
- Updated unit tests to validate the new configuration handling for approval and search features, ensuring accurate argument parsing.
These changes enhance the functionality of the CodexProvider, allowing for more dynamic command configurations and improving test coverage.
- Reorganized argument construction in CodexProvider to separate pre-execution arguments from global flags, improving clarity and maintainability.
- Updated unit tests to reflect changes in argument order, ensuring correct validation of approval and search indices.
These changes enhance the structure of the CodexProvider's command execution process and improve test reliability.
- Introduced a new method to check Codex authentication status, allowing for better handling of API keys and OAuth tokens.
- Updated API key management to include OpenAI, enabling users to manage their keys more effectively.
- Enhanced the CodexProvider to support session ID tracking and deduplication of text blocks in assistant messages.
- Improved error handling and logging in authentication routes, providing clearer feedback to users.
These changes improve the overall user experience and security of the Codex integration, ensuring smoother authentication processes and better management of API keys.
- Introduced a new method in ProviderFactory to check if a model supports vision/image input.
- Updated AgentService and AutoModeService to validate vision support before processing images, throwing an error if the model does not support it.
- Enhanced error messages to guide users on switching models or removing images if necessary.
These changes improve the robustness of image processing by ensuring compatibility with the selected models.
- Added optional API keys for OpenAI and Cursor to the .env.example file.
- Implemented API key validation in CursorProvider to ensure valid keys are used.
- Introduced rate limiting in Claude and Codex authentication routes to prevent abuse.
- Created secure environment handling for authentication without modifying process.env.
- Improved error handling and logging for authentication processes, enhancing user feedback.
These changes improve the security and reliability of the authentication mechanisms across the application.
- Updated CodexProvider to read prompts from stdin to prevent shell escaping issues.
- Enhanced AgentService to handle streamed error messages from providers, ensuring a consistent user experience.
- Modified UI components to display error messages clearly, including visual indicators for errors in chat bubbles.
- Updated CLI status handling to support both Claude and Codex APIs, improving compatibility and user feedback.
These changes enhance the robustness of the application and improve the user experience during error scenarios.
- Added a new function to retrieve subagents configuration from settings, allowing users to enable/disable subagents and select sources for loading them.
- Updated the AgentService to incorporate subagents configuration, dynamically adding tools based on the settings.
- Enhanced the UI components to manage subagents, including a settings section for enabling/disabling and selecting sources.
- Introduced a new hook for managing subagents settings state and interactions.
These changes improve the flexibility and usability of subagents within the application, enhancing user experience and configuration options.
- Updated .gitignore to include skills directory.
- Introduced agent discovery functionality to scan for AGENT.md files in user and project directories.
- Added new API endpoint for discovering filesystem agents.
- Implemented UI components for managing skills and viewing custom subagents.
- Enhanced settings helpers to retrieve skills configuration and custom subagents.
- Updated agent service to incorporate skills and subagents in task delegation.
These changes enhance the capabilities of the system by allowing users to define and manage skills and custom subagents effectively.
- Removed MCP permission settings from the application, including related functions and UI components.
- Updated SDK options to always bypass permissions and allow unrestricted tool access in autonomous mode.
- Adjusted related components and services to reflect the removal of MCP permission configurations, ensuring a cleaner and more efficient codebase.
- Add getCliPath() public method to CursorProvider to avoid private field access
- Add path validation to cursor-config routes to prevent traversal attacks
- Add supportsVision field to CursorModelConfig (all false - CLI limitation)
- Consolidate duplicate types in providers/types.ts (re-export from @automaker/types)
- Add MCP servers warning log instead of error (not yet supported by Cursor CLI)
- Fix debug log type safety (replace 'as any' with proper type narrowing)
- Update docs to remove non-existent tier field, add supportsVision field
- Remove outdated TODO comment in sdk-options.ts
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Integrated a centralized logging system using createLogger from @automaker/utils, replacing console.log and console.error statements with logger methods for consistent log formatting and improved readability.
- Updated various modules, including auth, events, and services, to utilize the new logging system, enhancing error tracking and operational visibility.
- Refactored logging messages to provide clearer context and information, ensuring better maintainability and debugging capabilities.
This update significantly enhances the observability of the server components, facilitating easier troubleshooting and monitoring.
- Introduced a new function, buildThinkingOptions, to handle the conversion of ThinkingLevel to maxThinkingTokens for the Claude SDK.
- Updated existing SDK option creation functions to incorporate thinking options, ensuring that maxThinkingTokens are included based on the specified thinking level.
- Enhanced the settings service to support migration of phase models to include thinking levels, improving compatibility with new configurations.
- Added comprehensive tests for thinking level integration and migration logic, ensuring robust functionality across the application.
This update significantly improves the SDK's configurability and performance by allowing for more nuanced control over reasoning capabilities.
- Removed the 'tier' property from Cursor model configurations and related UI components.
- Updated relevant files to reflect the removal of tier-related logic and display elements.
- This change simplifies the model structure and UI, focusing on essential attributes.
Carefully merged latest changes from main branch into the Cursor CLI integration
branch. This merge brings in important improvements and fixes while preserving
all Cursor-related functionality.
Key changes from main:
- Sandbox mode security improvements and cloud storage compatibility
- Version-based settings migrations (v2 schema)
- Port configuration centralization
- System paths utilities for CLI detection
- Enhanced error handling in HttpApiClient
- Windows MCP process cleanup fixes
- New validation and build commands
- GitHub issue templates and release process improvements
Resolved conflicts in:
- apps/server/src/routes/context/routes/describe-image.ts
(Combined Cursor provider routing with secure-fs imports)
- apps/server/src/services/auto-mode-service.ts
(Merged failure tracking with raw output logging)
- apps/server/tests/unit/services/terminal-service.test.ts
(Updated to async tests with systemPathExists mocking)
- libs/platform/src/index.ts
(Combined WSL utilities with system-paths exports)
- libs/types/src/settings.ts
(Merged DEFAULT_PHASE_MODELS with SETTINGS_VERSION constants)
All Cursor CLI integration features remain intact including:
- CursorProvider and CliProvider base class
- Phase-based model configuration
- Provider registry and factory patterns
- WSL support for Windows
- Model override UI components
- Cursor-specific settings and configurations
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- apps/server/src/providers/claude-provider.ts
Add ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN to the environment variable
allowlist, enabling use of LLM gateways (LiteLLM, Helicone) and Anthropic-
compatible providers (GLM 4.7, Minimax M2.1, etc.).
Closes#338
