Tier 1: Causal Reasoning Quick Wins - CoT Annotations, Root-Cause API, Causal Prompts

[mrveiss]

Summary

Implements foundational causal reasoning infrastructure for AutoBot: CoT event annotations, root-cause API, and causal reasoning guidance in agent prompts.

Changes

1. CoT Event Causal Annotations

• Added CausalLink dataclass for capturing why decisions were made
• All 6 emit_* functions accept optional causal_chain: List[CausalLink]
• Automatic sensitive data redaction
• 26 comprehensive tests (all passing)

Files:

• autobot-backend/chat_workflow/cot_events_test.py (390 lines, 26 tests)
• autobot-backend/reasoning/causal_reasoning.py (314 lines)

2. Root-Cause Analysis API Endpoint

• New service: RootCauseAnalyzer with confounder detection
• API endpoint: GET /api/analytics/root-cause/{task_id}
• Returns causal chain + confounders + explanations
• 23 tests (all passing)

Files:

• autobot-backend/services/root_cause_analyzer.py (383 lines)
• autobot-backend/api/analytics_root_cause_test.py (535 lines, 23 tests)

3. Causal Reasoning in Agent Prompts

• New CAUSAL_ANALYSIS category for Think Tool
• Guides LLMs to explain causality
• 40+ tests

Files:

• autobot-backend/orchestration/causal_error_analyzer.py (276 lines)
• autobot-backend/tests/agents/test_causal_reasoning.py (418 lines, 40+ tests)

Stats

• Lines: ~1,500
• Tests: 89 (all passing)
• Production-ready: Yes
• Backward compatible: Yes

Test Plan

• All 89 tests passing
• CoT events emit with causal context
• Root-cause API working correctly
• Agent prompts guide causal reasoning

🤖 Generated with Claude Code