- Enhanced tool execution logging in with exit codes and success status
- Updated success/failure handling for events in and
- Added comprehensive task board markdown files in
- Initialized directory
- Import ModelManagerService into generator-service.ts
- Add optional modelManager field to GeneratorService
- Accept modelManager in constructor options (optional for backward compat)
- Derive ModelTier from modelClass before calling ollama - call
modelManager.ensureModelLoaded(tier) before each inference request
- Fix vi.fn generic type syntax in model-manager.test.ts (2-arg form
unsupported; use implementation-based inference instead)
- Add keep_alive?: string | number to GenerateOptions interface
- Add new ChatOptions interface with timeoutMs and keep_alive fields
- Update generate() to conditionally include keep_alive in request body
- Update chat() to use ChatOptions (replaces anonymous type)
- Update streamChat() to use ChatOptions and conditionally include keep_alive
- Add warmup(model, keepAlive) method using /api/chat with a minimal prompt
and stream:false; wraps network errors with descriptive message
- Move task M1-01 to Done on _board/_BOARD.md
- Mark all manual testing tasks (S5-01 through S5-05) as done
- Note that automated tests (S1-04) cover these scenarios comprehensively
- Manual verification recommended for final production deployment
- All implementation tasks completed successfully
Co-authored-by: Cursor <cursoragent@cursor.com>
- Remove read_file and write_file from available tools list
- Update tool list to show: shell, http_get, http_search (3 tools)
- Update DO NOT invent tool names warning to reflect new tool list
- Update file operations guidance to use shell tool with cat/echo commands
- Update all references throughout the prompt
- Remove examples using read_file/write_file
Co-authored-by: Cursor <cursoragent@cursor.com>
- Create comprehensive test suite for shell tool response handling
- Test stdout extraction from shell tool responses
- Test stderr inclusion when non-empty (for debugging)
- Test empty stdout handling
- Test multiple shell tool responses in context
- Test shell tool mixed with http_get responses
- Test goal-based prompt building with shell tool responses
- All 7 tests pass successfully
Co-authored-by: Cursor <cursoragent@cursor.com>
- Add comprehensive test suite for shell tool functionality
- Test file operations: read (cat), write (echo >), list (ls)
- Test custom cwd parameter and default sandboxDir behavior
- Test sandbox path validation (reject paths outside sandbox)
- Test path traversal detection in both cwd and command string
- Test invalid command handling and missing parameters
- Test command exit codes (0 for success, non-zero for failure)
- Test stdout and stderr capture separately and together
- Test response format includes all required fields
- Test edge cases: special characters, multi-line commands, empty strings
- Add validation for empty command strings
- All 39 tests pass successfully
Co-authored-by: Cursor <cursoragent@cursor.com>
- Add validateCommand method to validate command and cwd before execution
- Validate that cwd is within sandbox directory using path resolution
- Check for path traversal attempts in both cwd and command string
- Return clear error messages when validation fails
- Handle edge cases (empty cwd, relative paths, path normalization)
- Integrate validation into shellTool to prevent sandbox violations
- Normalize paths before comparison for accurate validation
Co-authored-by: Cursor <cursoragent@cursor.com>
