Skip to content

Latest commit

 

History

History
41 lines (32 loc) · 1.34 KB

ARCHITECTURE.md

File metadata and controls

41 lines (32 loc) · 1.34 KB

Architecture

Trinity is an executable control plane for AI-assisted work.

Human Owner
    |
    v
Trinity Kernel
    |
    +-- Planner        (turn intent into scoped plan artifacts)
    +-- Executor       (run bounded work in approved scope)
    +-- Verifier       (check evidence and decide PASS/RETRY/NEEDS_HUMAN/DEAD)
    +-- Audit Chain    (append-only truth log)
    +-- Capture Store  (inputs, outputs, and evidence)
    +-- Memory Index   (retrieval over accepted artifacts)

Core Principle

Trinity does not trust claims. Trinity trusts artifacts.

An AI agent can propose, edit, and produce evidence. It cannot self-approve completion, bypass policy gates, or turn memory into truth.

Runtime Layers

  • CLI kernel: deterministic commands in .ai/cli/.
  • Session capsule: per-task working area under .ai/sessions/<id>/.
  • State pointer: .state/session_state.json, with canonical graph_state and compatibility-only legacy_state.
  • Audit chain: .ai/audit/events.ndjson, append-only and hash-linked.
  • Verifier: rule-driven checks that consume artifacts.
  • Memory: retrieval index over accepted artifacts; rrr uses memory-cli index.

Canonical Specs

Detailed contracts live under docs/specs/. Operator docs are practical entry points; the specs remain canonical for implementation details.