Infer relationship between interceptions to build a causal chain

[dannykopping]

I have a draft RFC and POC for this in the works; I'll add more details later.

Problem: we're currently unable to trace causal relationships between human prompts and AI agent actions, which reduces the UX of the audit experience they can't see how A (human prompt)-> B (agentic loop entered) -> ... -> Z (database dropped).

AI Bridge currently treats each API interception as logically separate because the underlying APIs (/v1/messages, /v1/chat/completions) are stateless. This creates critical problems for attribution and auditing:

1. No visible relationships: When an AI agent enters an agentic loop (making multiple tool calls to satisfy a request), we display 5+ separate interceptions with no indication they're related
2. Loss of context: Subsequent interceptions appear to have empty prompts, making it unclear what they represent
3. Attribution gaps: Cannot distinguish human-initiated actions from agent-initiated ones
4. Audit failures: Auditors cannot trace the chain of events from initial human intent through to agent actions, which is essential for determining liability and policy violations

In the age of AI collaboration, accurate provenance is becoming increasingly critical - only humans can be held liable, so we need clear evidence of who initiated what.

Current state: Interceptions displayed in flat reverse-chronological list

Needed: Hierarchical view showing threads of human-agent interaction with clear parent/child relationships