Adr¶
Table of Contents:
ADR-OA01: TNH-Conductor — Provenance-Driven AI Workflow Coordination - Strategic architecture for coordinating external AI agents (Claude Code, Codex) through bounded, auditable, human-supervised workflows
ADR-OA01.1: TNH-Conductor — Provenance-Driven AI Workflow Coordination (v2) - Strategic architecture for coordinating external AI agents through bounded, auditable, human-supervised workflows with CLI opcode tooling
ADR-OA01.2: Conceptual Spike for Orientation-Based Supervisory Orchestration - Strategy-level spike direction for testing orientation-based supervisory orchestration as an alternative to engine-first workflow control.
ADR-OA01.3: Practical Approach for the Orientation-Based Conceptual Spike - Practical approach for testing the OA01.2 orientation-based collaboration hypothesis with minimal new code and explicit prerequisite spike work.
ADR-OA01.4: Headless Agent Communication Functional Spike - Prerequisite functional spike for proving that structured headless communication between supervising and worker agents can happen simply and repeatably enough to support the OA01.2 conceptual spike.
ADR-OA02: Phase 0 Protocol Layer Spike — Headless Capture + Safety Controls - De-risking spike to prove headless agent capture and safety controls for tnh-conductor
ADR-OA03: Agent Runner Architecture - Phase 1 architecture for agent execution — kernel + adapter pattern based on OA02 spike learnings
ADR-OA03.1: Claude Code Runner - Claude Code execution path — headless print mode only, no interactive/PTY automation
ADR-OA03.2: Codex Runner - Codex execution path — API-first, tool-driven runner for implementation tasks
ADR-OA03.3: Codex CLI Runner - Codex execution path via CLI — headless exec mode, superseding API-based approach
ADR-OA04: Workflow Execution Contracts - Defines the canonical workflow document format, opcode semantics, and execution-contract foundation for tnh-conductor runtime implementation.
ADR-OA04.1: MVP Runtime Build-Out Sequence - Implementation-guide addendum for OA04 defining the MVP runtime build order, role split, and handoff to OA07 bootstrap workspace work.
ADR-OA04.2: Runner Contract - Defines the maintained RUN_AGENT request/result, artifact, and normalization contract for Claude CLI and Codex CLI adapters.
ADR-OA04.3: Provenance and Run-Artifact Contract - Defines the maintained run directory, artifact manifest, and event/provenance handoff contract for workflow execution.
ADR-OA04.3.1: Run Transparency and State Reporting - Adds a maintained operator-facing status contract on top of the OA04.3 run-artifact model for live monitoring of headless workflow runs.
ADR-OA04.4: Policy Enforcement Contract - Defines the shared permissibility model, native-control mapping boundary, and violation handling contract for workflow execution.
ADR-OA04.5: Harness Backend Contract - Defines the maintained backend contract for executing generated and predefined validation harnesses under RUN_VALIDATION.
ADR-OA05: Prompt Library Specification - Defines prompt artifact schema, versioning, rendering, and catalog validation contracts for tnh-conductor prompt-program behavior.
ADR-OA06: Planner Evaluator Contract - Defines planner evaluator I/O schemas, status derivation, contradiction rules, and deterministic decision vectors for EVALUATE steps.
ADR-OA06.1: Evaluator-Directed Revision Loop - Defines a bounded maintained review-to-revision loop for conductor workflows so review and validation artifacts can drive one deterministic refinement pass.
ADR-OA07: Diff-Policy + Safety Rails - Defines the bootstrap safety model for worktree isolation, branch-scoped rollback, and agent authority boundaries in maintained agent orchestration.
ADR-OA07.1: Worktree Lifecycle and Rollback - Bootstrap implementation guide for managed worktree creation, run-directory separation, and branch-scoped rollback in maintained agent orchestration.
This file auto-generated.