# Error Handling Patterns Reference
**Source:** `docs/THREAD_AND_STREAMING_ARCHITECTURE.md` Appendix A.8
## Tool Chain Error Wrapping
Each execution layer catches errors and rethrows with context:
```python
@dataclass
class ToolChainError(Exception):
code: str
message: str
chain: List[str] # ["anthropic_thread", "anthropic_messages", "http_client"]
failed_at: FailedToolContext
cause: Optional[Exception]
```
## Error Response Format
```json
{
"success": false,
"error": {
"code": "TOOL_CHAIN_FAILED",
"message": "Tool chain failed at 'anthropic_messages': invalid config.",
"chain": ["anthropic_thread", "anthropic_messages", "http_client"],
"failed_at": {
"tool_id": "anthropic_messages",
"config_path": ".ai/tools/llm/anthropic_messages.yaml",
"validation_errors": [...]
}
}
}
```
## Proactive Validation
- Validate tool YAML configs at load/registration time
- Mark tools "unavailable" with clear diagnostic
- Avoids confusing runtime failures mid-thread
## Stream Error Recovery
- Completed tool calls ARE executed
- Partial tool calls are discarded (not executed)
- Mark stream as incomplete, return error with context
See doc Appendix A.3 for stream parsing error recovery details.
