Implement a complete API Management Interface for React frontend with split-
panel layout, migrating all features from legacy JS frontend.
New Features:
- API Settings page with 5 tabs: Providers, Endpoints, Cache, Model Pools, CLI Settings
- Provider Management: CRUD operations, multi-key rotation, health checks, test connection
- Endpoint Management: CRUD operations, cache strategy configuration, enable/disable toggle
- Cache Settings: Global configuration, statistics display, clear cache functionality
- Model Pool Management: CRUD operations, auto-discovery feature, provider exclusion
- CLI Settings Management: Provider-based and Direct modes, full CRUD support
- Multi-Key Settings Modal: Manage API keys with rotation strategies and weights
- Manage Models Modal: View and manage models per provider (LLM and Embedding)
- Sync to CodexLens: Integration handler for provider configuration sync
Technical Implementation:
- Created 12 new React components in components/api-settings/
- Extended lib/api.ts with 460+ lines of API client functions
- Created hooks/useApiSettings.ts with 772 lines of TanStack Query hooks
- Added RadioGroup UI component for form selections
- Implemented unified error handling with useNotifications across all operations
- Complete i18n support (500+ keys in English and Chinese)
- Route integration (/api-settings) and sidebar navigation
Code Quality:
- All acceptance criteria from plan.json verified
- Code review passed with Gemini (all 7 IMPL tasks complete)
- Follows existing patterns: Shadcn UI, TanStack Query, react-intl, Lucide icons
- Fix prompt incorrectly passed to codex review with target flags (--uncommitted/--base/--commit)
- Remove unsupported --skip-git-repo-check flag from codex review command
- Fix TypeScript type error in codex-lens.ts (missing 'installed' property)
Codex CLI constraint: --uncommitted/--base/--commit and [PROMPT] are mutually exclusive.
- Implemented hooks for CRUD operations on providers and endpoints.
- Added cache management hooks for cache stats and settings.
- Introduced model pool management hooks for high availability and load balancing.
- Created localization files for English and Chinese translations of API settings.
- Implemented E2E tests for workspace switching functionality, covering scenarios such as switching workspaces, data isolation, language preference maintenance, and UI updates.
- Added tests to ensure workspace data is cleared on logout and handles unsaved changes during workspace switches.
- Created comprehensive backend tests for the ask_question tool, validating question creation, execution, answer handling, cancellation, and timeout scenarios.
- Included edge case tests to ensure robustness against duplicate questions and invalid answers.
- Added A2UITypes for defining question structures and answers.
- Created A2UIWebSocketHandler for managing WebSocket connections and message handling.
- Developed ask-question tool for interactive user questions via A2UI.
- Introduced platformUtils for platform detection and shell command handling.
- Centralized TypeScript types in index.ts for better organization.
- Implemented compatibility checks for hook templates based on platform requirements.
- Enhance CLI status rendering to display settings file information for builtin Claude.
- Introduce settings file input in CLI manager for configuring the path to settings.json.
- Update Claude CLI tool interface to include settingsFile property.
- Implement settings file resolution and validation in CLI executor.
- Create a new collaborative planning workflow command with detailed documentation.
- Add test scripts for debugging tool configuration and command building.
- Deleted obsolete INDEX.md and OPTIMIZATION_SUMMARY.md files, consolidating documentation for improved clarity and organization.
- Removed skipGitRepoCheck option from CLI execution parameters to streamline command usage.
- Updated CLI executor utilities to automatically skip git repository checks, allowing execution in non-git directories.
- Enhanced documentation with new ARCHITECTURE.md and INDEX.md files for better navigation and understanding of the system architecture.
- Created CONTENT_MIGRATION_REPORT.md to verify zero content loss during the consolidation process.
Problem: Codex CLI uses --json flag to output JSONL events, but executor was using plain text parser. This prevented proper parsing of structured events, breaking session creation.
Root cause: buildCommand() added --json flag for Codex but never communicated this to the output parser. Result: JSONL events treated as raw text → session markers lost.
Solution:
- Extend buildCommand() to return outputFormat
- Auto-detect 'json-lines' when tool is 'codex'
- Use auto-detected format in executeCliTool()
- Properly parse structured events and extract session data
Files modified:
- ccw/src/tools/cli-executor-utils.ts: Add output format auto-detection
- ccw/src/tools/cli-executor-core.ts: Use auto-detected format for parser
- ccw/src/commands/cli.ts: Add debug instrumentation
Verified:
- Codex outputs valid JSONL (confirmed via direct test)
- CLI_EXECUTION_STARTED events broadcast correctly
- Issue was downstream in output parsing, not event transmission
- Updated Phase 1 and Phase 2 documentation to include next phase links and data flow details.
- Expanded Phase 5 documentation to include comprehensive validation and README generation steps, along with validation report structure.
- Added purpose and usage context sections to various action and script templates (e.g., autonomous-action, llm-action, script-bash).
- Improved commands management by simplifying the command scanning logic and enabling/disabling commands through renaming files.
- Enhanced dashboard command manager to format group names and display nested groups with appropriate icons and colors.
- Updated LiteLLM executor to allow model overrides during execution.
- Added action reference guide and template reference sections to the skill-tuning SKILL.md for better navigation and understanding.
- Implemented commands routes for listing, enabling, and disabling commands.
- Created commands manager view with accordion groups for better organization.
- Added loading states and confirmation dialogs for enabling/disabling commands.
- Enhanced error handling and user feedback for command operations.
- Introduced CSS styles for commands manager UI components.
- Updated navigation to include commands manager link.
- Refactored existing code for better maintainability and clarity.
- Added new API endpoints to enable and disable skills.
- Introduced logic to manage disabled skills, including loading and saving configurations.
- Enhanced skills routes to return lists of disabled skills.
- Updated frontend to display disabled skills and allow toggling their status.
- Added internationalization support for new skill status messages.
- Created JSON schemas for plan verification agent and findings.
- Defined new types for skill management in TypeScript.
- Updated `plan.md` to include new fields in context-package.json: prioritized_context, user_intent, priority_tiers, dependency_order, and sorting_rationale.
- Added validation for the existence of the prioritized_context field in context-package.json.
- Modified user decision flow in task generation to present action choices after planning completion.
- Improved context-gathering process in `context-gather.md` to integrate user intent and prioritize context based on user goals.
- Revised conflict-resolution documentation to require planning notes records after conflict analysis.
- Streamlined task generation in `task-generate-agent.md` to utilize pre-sorted context without redundant sorting.
- Removed unused settings persistence functions and corresponding tests from `claude-cli-tools.ts` and `settings-persistence.test.ts`.
- Removed deprecated constants and functions from cli-config-manager.ts.
- Introduced new provider model presets in litellm-provider-models.ts for better organization and management of model information.
- Created provider-routes.ts to handle API endpoints for retrieving provider information and models.
- Added integration tests for provider routes to ensure correct functionality and response structure.
- Implemented unit tests for settings persistence functions, covering various scenarios and edge cases.
- Enhanced error handling and validation in the new routes and settings functions.
- Implemented the ccw command as a main workflow orchestrator.
- Added a 5-phase workflow including intent analysis, requirement clarification, workflow selection, user confirmation, and command execution.
- Developed functions for analyzing user input, selecting workflows, and executing command chains.
- Integrated TODO tracking for command execution progress.
- Created comprehensive tests for the CommandRegistry, covering YAML parsing, command retrieval, and error handling.
- Remove hardcoded Windows paths (D:\Claude_dms3\codex-lens) that were displayed to macOS/Linux users
- Generate dynamic possible paths list based on runtime environment
- Support multiple installation locations (cwd, project root, home directory)
- Improve error messages with platform-appropriate paths
- Maintain consistency across both bootstrapWithUv() and installSemanticWithUv() functions
Fixes remaining issue from #104 regarding cross-platform error message compatibility
- Implement two-pass search strategy for codex-lens path detection
- First pass: prefer non-node_modules paths (development environment)
- Second pass: allow node_modules paths (NPM global install)
- Fixes CodexLens installation for all NPM global install users
- No breaking changes, maintains backward compatibility
Resolves issue where NPM global install users could not install CodexLens
because the code rejected paths containing /node_modules/, which is the
only valid location for codex-lens in NPM installations.
Tested on macOS with Node.js v22.18.0 via NPM global install.
Changed loadClaudeCliTools() to only load tools explicitly defined
in user config. Previously, DEFAULT_TOOLS_CONFIG.tools was spread
before user tools, causing all default tools to be loaded even if
not present in user config.
User config now has complete control over which tools are loaded.
The ensureToolTags() function was only returning enabled, primaryModel,
secondaryModel, and tags - missing envFile. This caused envFile to be
lost during config merge in loadClaudeCliTools().
Related to #96 - gemini envFile setting lost after page refresh
Fixes#96 - gemini/qwen envFile setting was lost after page refresh
because getFullConfigResponse() was not including the envFile field
when converting config to the legacy API format.
Changes:
- Add envFile?: string | null to CliToolConfig interface
- Include envFile in getFullConfigResponse() conversion
Fix task click handlers not working in multi-CLI planning detail page.
Root cause: liteTaskDataStore was not being populated with multiCliPlan
sessions during initialization, so task click handlers couldn't access
session data using currentSessionDetailKey.
Changes:
- navigation.js: Add code to populate multiCliPlan sessions in liteTaskDataStore
- notifications.js: Add code to populate multiCliPlan sessions when data refreshes
Now when task detail page loads, liteTaskDataStore contains the correct key
'multi-cli-${sessionId}' matching currentSessionDetailKey, allowing task
click handlers to find session data and open detail drawer.
Verified: Task clicks now properly open detail panel for all 7 tasks.
- Fix missing i18n translations: loop.add, loop.save, loop.cancel
- Replace hardcoded validTools with dynamic tool loading from cli-tools.json
- Support external CLI wrappers (like doubao) in task creation and updates
- Add getEnabledToolsList() helper to fetch enabled tools dynamically
- Update mapIssueToolToLoopTool() to accept any string tool name
- Update validateTool() to use dynamic tool list
- Change LoopTask.tool type from specific strings to string (accepts any tool)
This allows tasks to use any enabled CLI tool from configuration,
including builtin tools, cli-wrappers, and api-endpoints, not just
the hardcoded ['bash', 'gemini', 'codex', 'qwen', 'claude'].
- Implemented loop control tasks in JSON format for testing.
- Created comprehensive test scripts for loop flow and standalone tests.
- Developed a shell script to automate the testing of the entire loop system flow, including mock endpoints and state transitions.
- Added error handling and execution history tests to ensure robustness.
- Established variable substitution and success condition evaluations in tests.
- Set up cleanup and workspace management for test environments.
- Updated package name from `codexlens` to `codex-lens` in all relevant files to ensure consistency with `pyproject.toml`.
- Enhanced `findLocalPackagePath()` to always search for local paths, even when running from `node_modules`.
- Removed fallback logic for PyPI installation in several functions, providing clearer error messages for local installation failures.
- Added detailed documentation on installation steps and error handling for local development packages.
- Introduced a new summary document outlining the issues and fixes related to CodexLens installation.
- Created a comprehensive guide for setting up TypeScript LSP in Claude Code, detailing installation methods, configuration, and troubleshooting.
- Added multiple debugging test scripts to validate LSP communication with pyright, including direct communication tests, configuration checks, and document symbol retrieval.
- Implemented error handling and logging for better visibility during LSP interactions.
- Introduced a new test script for the CodexLens LSP tool to validate core functionalities including symbol search, find definition, find references, and get hover.
- Created comprehensive documentation for the MCP endpoint design, detailing the architecture, features, and integration with the CCW MCP Manager.
- Developed a detailed implementation plan for transitioning to a real LSP server, outlining phases, architecture, and acceptance criteria.
Introduce dedicated agent_message IR type to distinguish final AI responses
from generic stdout. This enables --final flag to show only agent messages,
filtering out all intermediate content (JSONL events, reasoning, tool calls).
Changes:
- Add agent_message type to CliOutputUnitType
- Update JsonLinesParser to map final responses from all tools (codex,
gemini, claude, opencode) to agent_message type
- Add final_output field to database schema with migration
- Update getCachedOutput and getConversation to return finalOutput
- Prefer finalOutput in outputAction for --final flag
Fixes issue where --final showed raw JSONL instead of filtered content.
- Skip template concatenation when using --uncommitted/--base/--commit
- Allow empty prompt for review mode with target flags
- Add hasReviewTarget check in command routing
- Update documentation with validation constraints
codex review constraint: target flags and prompt are mutually exclusive
- Replace $PROTO/$TMPL environment variable injection with systemRules/roles direct concatenation
- Append rules to END of prompt instead of prepending
- Change prompt field name from RULES to CONSTRAINTS in all prompts
- Default to universal-rigorous-style template when --rule not specified
- Update all .claude documentation, agents, commands, and skills
- Add streaming_content type support for Gemini delta messages
Breaking: Prompts now use CONSTRAINTS field instead of RULES
- Add 'review' to mode enum in ParamsSchema and schema
- Implement codex review subcommand in buildCommand (uses --uncommitted by default)
- Other tools (gemini/qwen/claude) accept review mode but no operation change
- Update cli-tools-usage.md with review mode documentation
- Added loading of plan.json in scanMultiCliDir to improve task extraction.
- Implemented normalization of tasks from plan.json format to support new UI.
- Updated CSS for multi-CLI plan summary and task item badges for better visibility.
- Refactored hook-manager to use Node.js for cross-platform compatibility in command execution.
- Improved i18n support for new CLI tool configuration in the hook wizard.
- Enhanced lite-tasks view to utilize normalized tasks and provide better fallback mechanisms.
- Updated memory-update-queue to return string messages for better integration with hooks.
- Introduced a new input group for environment file configuration in the dashboard CSS.
- Updated hook manager to queue CLAUDE.md updates with configurable threshold and timeout.
- Enhanced CLI manager view to include environment file input for built-in tools (gemini, qwen).
- Implemented environment file loading mechanism in cli-executor-core, allowing custom environment variables.
- Added unit tests for environment file parsing and loading functionalities.
- Updated memory update queue to support dynamic configuration of threshold and timeout settings.
- Introduced `/issue:discover-by-prompt` command for user-driven issue discovery.
- Implemented multi-agent exploration with iterative feedback loops.
- Added ACE semantic search for context gathering and cross-module comparison capabilities.
- Enhanced user experience with natural language input and adaptive exploration strategies.
feat: implement memory update queue tool for batching updates
- Created `memory-update-queue.js` for managing CLAUDE.md updates.
- Added functionality for queuing paths, deduplication, and auto-flushing based on thresholds and timeouts.
- Implemented methods for queue status retrieval, flushing, and timeout checks.
- Configured to store queue data persistently in `~/.claude/.memory-queue.json`.
- #70: Fix API Key Tester URL handling - normalize trailing slashes before
version suffix detection to prevent double-slash URLs like //models
- #69: Fix memory embedder ignoring CodexLens config - add error handling
for CodexLensConfig.load() with fallback to defaults
- #68: Fix ccw cli using wrong Python environment - add getCodexLensVenvPython()
to resolve correct venv path on Windows/Unix
- #67: Fix LiteLLM API Provider test endpoint - actually test API key connection
instead of just checking ccw-litellm installation
- #66: Fix help-routes.ts path configuration - use correct 'ccw-help' directory
name and refactor getIndexDir to pure function
- #63: Fix CodexLens install state refresh - add cache invalidation after
config save in codexlens-manager.js
Also includes targeted unit tests for the URL normalization logic.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
