Repository navigation
EN Home
한국어 | English
Agent Governance Suite (AGS) is a local governance layer for coordinating, validating, and carrying forward AI work across Codex and Claude Code sessions.
The repository README is the fastest way to install AGS and try it. This Wiki focuses on the concepts, architecture, collaboration model, and governance mechanisms behind the system.
- What AGS Is — and Is Not
- Core Concepts
- Architecture
- Peer Sessions
- Getting Started
- Messaging and ACK
- Governance and Evidence
A long-running AI task is difficult to manage even in a single session. Goals drift, scope expands, earlier decisions become detached from the current state, and "done" can become a claim rather than a verified result.
Using multiple sessions adds another layer of operational problems:
- Who owns which part of the work?
- Which session currently holds the relevant context?
- How does one session send conditions, evidence, or review results to another?
- Was a message merely delivered, or was the requested work actually completed?
- Can an implementation be reviewed independently?
- Can a task survive interruption, compaction, or session changes?
- Is completion tied to current evidence rather than stale assertions?
AGS addresses these problems through a set of coordinated mechanisms.
- Task contracts define goals, scope, completion criteria, risk, and authority.
- Specialist skills provide task-specific decision procedures and output contracts.
- Independent sessions retain their own context while collaborating through PEER messages.
- The session board makes active responsibilities discoverable.
- The local MCP and SQLite runtime records workflow state, stage results, evidence, and completion gates.
- Independent audit and evidence validation separate "work was attempted" from "completion was demonstrated."
AGS is not primarily a framework for spawning subagents.
Its core concern is governing how independent AI sessions coordinate, preserve responsibility, exchange evidence, audit one another, and continue long-running work.
Agent creation can be part of the system, but it is not the organizing principle.
See What AGS Is — and Is Not for the full distinction.
| Location | Purpose |
|---|---|
| Repository README | Product overview, installation, quick demo |
| Wiki | Concepts, operating model, architecture, design rationale |
docs/ |
Implementation details, protocols, operations, release-specific technical material |
skills/ |
Specialist skill contracts and execution guidance |