feat: implement H3ERE-compliant MCP Bridge & Audit Anchoring

[phahim1]

Phase 1: H3ERE Governance Implementation

This PR delivers the core alignment and communication infrastructure for the CIRISNode.

Key Technical Deliverables:

MCP Bridge Integration: Orchestrates local Gemma 3 (via Ollama) to perform complex logic puzzles with full Chain-of-Thought (CoT) reasoning.

Requirement #4 (Cryptographic Audit): Implemented an immutable Decision Ledger using raw SQL anchoring to SQLite. Every AI action is now timestamped and recorded for governance.

Automated Result Extraction: Integrated a robust Regex grading engine to extract final conclusions from long-form reasoning traces, ensuring accurate alignment scoring.

Code Optimization: Refactored the benchmark core from ~300 lines down to a lean 166-line stable production build.

Verification Status:

POST /api/v1/simplebench/run-sync verified via Swagger.

Audit persistence verified via /api/v1/audit/logs.

[sonarqubecloud]

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[emooreatx]

Hey @phahim1 - thanks for this contribution! We can see you're working toward AgentBeats Phase 1 requirements, and that's actually aligned with where we're heading with CIRISNode as the "green agent" / evaluator interface for HE-300.

Context You May Have Missed

We just merged PR #26 a few hours ago which added the EthicsEngine Enterprise (EEE) integration - this connects CIRISNode to our backend evaluation engine (now consolidated in CIRISBench). Unfortunately, this PR overwrites that work rather than building on it.

What We Like

• Using settings.JWT_SECRET instead of hardcoded secret - good catch!
• The audit anchoring concept aligns with AgentBeats' reproducibility requirements
• The regex-based answer grading improvements

How to Move Forward

We'd love to incorporate your hackathon work, but as additive changes rather than replacing the EEE integration. Here's what would help:

1. Rebase on current main - this has the EEE integration
2. Add audit logging alongside existing endpoints, don't replace them
3. Keep both provider options (OpenAI + Ollama) - AgentBeats requires A2A compatibility with various agents

AgentBeats Alignment

We're building CIRISNode as the green agent interface where purple agents connect to take HE-300. The architecture is:

Purple Agent (under test) → CIRISNode (green agent API) → CIRISBench (evaluation engine)

Your audit anchoring idea could help with AgentBeats' "reproducibility" requirement ("consistent results across runs"). Let's figure out how to add it properly.

Happy to chat more - feel free to join the discussion or ping us here!