Merge pull request #46 from hobbyistlabs-coder/feat/error-categorization-logging-10686897873219792435

feat: Implement error categorization for agent loop observability

AGENT_LOOP_IMPROVEMENTS.md

@@ -33,7 +33,7 @@ This document outlines a series of tasks to improve the main loop of the agentic
 ## Phase 5: Observability & Logging Dashboard
 
 * [ ] **Session Replay:** Capture the full "Chain of Thought" (CoT), including tool-call inputs/outputs, and state transitions to facilitate deep-dive session reviews.
-* [ ] **Error Categorization:** Distinguish between Model Failures (hallucinations/refusals), Infrastructure Failures (API timeouts), and Logic Failures (code execution errors).
+* [x] **Error Categorization:** Distinguish between Model Failures (hallucinations/refusals), Infrastructure Failures (API timeouts), and Logic Failures (code execution errors).
 * [ ] **Real-time Visuals:** Implement a lightweight UI/Dashboard using a Go-compatible framework like a custom Bubble Tea TUI to monitor the agent's health.
 
 ---

pkg/agent/loop_execute_llm.go

@@ -142,11 +142,19 @@ func (al *AgentLoop) executeLLMWithRetry(
 	}
 
 	if err != nil {
+		errorCategory := "infrastructure_failure"
+		if failErr := providers.ClassifyError(err, agent.ID, activeModel); failErr != nil {
+			if failErr.Reason == providers.FailoverFormat || failErr.Reason == providers.FailoverContextLength {
+				errorCategory = "model_failure"
+			}
+		}
+
 		logger.ErrorCF("agent", "LLM call failed",
 			map[string]any{
 				"agent_id":       agent.ID,
 				"iteration":      iteration,
 				"error":          err.Error(),
+				"error_category": errorCategory,
 			})
 		return nil, fmt.Errorf("LLM call failed after retries: %w", err)
 	}

pkg/agent/loop_llm.go

@@ -361,6 +361,15 @@ func (al *AgentLoop) runLLMIteration(
 				contentForLLM = r.result.Err.Error()
 			}
 
+			if r.result.IsError {
+				logger.ErrorCF("agent", "Tool execution failed",
+					map[string]any{
+						"tool":           r.tc.Name,
+						"error":          contentForLLM,
+						"error_category": "logic_failure",
+					})
+			}
+
 			toolResultMsg := providers.Message{
 				Role:       "tool",
 				Content:    contentForLLM,