926df2cfaf
- Complete architectural overhaul of useUnifiedCompletion hook - Unified state management: 8 separate states → single CompletionState interface - Simplified core logic: getWordAtCursor 194 lines → 42 lines (78% reduction) - Fixed infinite React update loops with ref-based input tracking - Smart triggering mechanism replacing aggressive auto-completion - Integrated @agent and @file mention system with system reminders - Added comprehensive agent loading and mention processing - Enhanced Tab/Arrow/Enter key handling with clean event management - Maintained 100% functional compatibility across all completion types Key improvements: • File path completion (relative, absolute, ~expansion, @references) • Slash command completion (/help, /model, etc.) • Agent completion (@agent-xxx with intelligent descriptions) • System command completion (PATH scanning with fallback) • Terminal-style Tab cycling, Enter confirmation, Escape cancellation • Preview mode with boundary calculation • History navigation compatibility • Empty directory handling with user feedback Architecture: Event-driven @mention detection → system reminder injection → LLM tool usage Performance: Eliminated 7-layer nested conditionals, reduced state synchronization issues Reliability: Fixed maximum update depth exceeded warnings, stable state management
58 lines
2.0 KiB
Markdown
58 lines
2.0 KiB
Markdown
# Agent Loop Fix Summary
|
|
|
|
## Root Cause Analysis
|
|
|
|
The agent loop was breaking because we introduced complex async @mention processing that interrupted the clean message flow. The original working codebase uses a simple, synchronous approach.
|
|
|
|
## Key Problems Identified
|
|
|
|
1. **Complex Async Processing**: The modified code added multiple async imports and checks during message processing
|
|
2. **Input Modification**: Removing @mentions and injecting system reminders changed the message structure
|
|
3. **Flow Interruption**: The async operations and input modifications broke the continuous agent execution loop
|
|
|
|
## Solution Applied
|
|
|
|
Restored the original simple message processing:
|
|
- Use `resolveFileReferences` to embed file content directly (as the original does)
|
|
- Remove complex agent mention detection
|
|
- Keep the message flow synchronous and clean
|
|
|
|
## Original vs Modified
|
|
|
|
### Original (Working):
|
|
```typescript
|
|
// Simple and direct
|
|
if (input.includes('@')) {
|
|
processedInput = await resolveFileReferences(processedInput)
|
|
}
|
|
```
|
|
|
|
### Modified (Broken):
|
|
```typescript
|
|
// Complex with multiple async operations
|
|
if (input.includes('@')) {
|
|
// Multiple async imports
|
|
// Agent detection logic
|
|
// System reminder generation
|
|
// Input modification
|
|
// ... 70+ lines of complex logic
|
|
}
|
|
```
|
|
|
|
## Why This Fixes the Issue
|
|
|
|
1. **Preserves Message Integrity**: Messages flow through without modification
|
|
2. **No Async Interruptions**: Simple, predictable execution
|
|
3. **Maintains Agent Context**: The agent loop can continue without context loss
|
|
|
|
## Testing
|
|
|
|
The agent loop should now work properly:
|
|
- Messages will process continuously
|
|
- Agents can execute multiple rounds
|
|
- @file mentions will embed content (as designed)
|
|
- No unexpected interruptions
|
|
|
|
## Lesson Learned
|
|
|
|
**Keep the message processing pipeline simple and synchronous.** Complex async operations and input modifications during message processing break the agent execution loop. Any special handling should be done at the tool execution level, not during message preparation. |