---
title: "okf-agent-graph"
description: "OKF Agent Graph Engineering Runtime (AGER): author multi-agent loop graphs as OKF and reverse-engineer existing agent stacks. Multi-host bindings and write isolation."
canonical: https://agentpluginsdirectory.com/plugins/okf-agent-graph
last-updated: 2026-10-02
---

# okf-agent-graph
OKF Agent Graph Engineering Runtime (AGER): author multi-agent loop graphs as OKF and reverse-engineer existing agent stacks. Multi-host bindings and write isolation.
- Slug: okf-agent-graph
- Publisher: Rick Hightower
- Repository: https://github.com/SpillwaveSolutions/okf-agent-graph
- Manifest: plugin.json
- Version: 0.8.1
- License: MIT
- Category (editorial): agent-tooling
- Skills: 8 (ager-author, ager-compile, ager-init-graph, ager-pack, ager-reverse-engineer, ager-scan, ager-session, ager-validate)
- MCP servers: 0
- Stars: 1
- Repository created: 2026-08-04
- Repository last pushed: 2026-08-30
- Publisher type: Organization
- Listing: https://agentpluginsdirectory.com/plugins/okf-agent-graph
- Schema: https://agent-plugins.org/schemas/1.0.0/plugin.schema.json

## What okf-agent-graph does, in the publisher's words

OKF Agent Graph Engineering Runtime (AGER): a Claude Code, Codex, Grok Build, Grok Bot, and LangChain Deep Agents plugin for authoring multi-agent loop graphs as OKF (Markdown + YAML).

Write isolation: docs/ISOLATION.md. LoopPolicy / KnowledgeBind read main and write brain/ /. Public examples use fictional lumenfield-detector and northstar-console only.

Also: Onboarding · skill ager-session.

- Orchestrator / Doer / Judge / Synthesizer roles
- LoopControls: goal, deadline, price, max_turns, no_progress
- Tools with expression block rules, idempotency, secrets
- ScratchPad KV (set / append lists) + lineage
- Run + Trigger, FailurePolicy, RetryPolicy, Compensation
- KnowledgeBind / RetrievalBinding (OKF / project knowledge)

From the project README, punctuation lightly normalized. Full text: https://raw.githubusercontent.com/SpillwaveSolutions/okf-agent-graph/HEAD/README.md

## Skills

- ager-author: Author or update AGER concepts: OrchestratorAgent, WorkerAgent, JudgeAgent, LoopPolicy/LoopControl, Tool+ToolRule, ScratchPad, Run/Trigger, FailurePolicy, KnowledgeBind. Use when editing multi-agent graphs, budgets, tool rules, or ops policies. Requires okf-plugin for link validation.
- ager-compile: Map an AGER AgentGraph to a target framework (LangGraph, CrewAI, OpenAI Agents, Anthropic lead/subagents) and emit adapter notes or stub skeletons. Not a full codegen runtime.
- ager-init-graph: Scaffold an OKF Agent Graph (AGER) bundle with orchestrator/workers/judge layout, LoopPolicy, ScratchPad, FailurePolicy, and sample Tool. Use when starting multi-agent graphs, AGER configs, or loop-engineering scaffolds. Requires okf-plugin (okf-graph-eng).
- ager-pack: Build a progressive-disclosure ContextPack from an AGER bundle. Default 1/4 window token budget, fail-closed. Bodies off unless that node is the pack root.
- ager-reverse-engineer: End-to-end reverse engineer of agent codebases (Claude/ChatGPT APIs, LangChain/LangGraph, CrewAI, LlamaIndex, Claude Agent SDK, Deep Agents, MCP, AgentCore, microVMs) into a draft AGER OKF knowledge graph. Primary reverse entry skill for okf-agent-graph (AGKC).
- ager-scan: Scan a codebase for agent frameworks (LangGraph, CrewAI, LlamaIndex, Claude/OpenAI APIs, MCP, Deep Agents, hyperscalers, microVMs) and for Claude/Grok/Codex/universal plugin agents and skills. Extracts prompts, tools, schemas, orchestration, loops, harness signals, and plugin frontmatter. Use when…
- ager-session: Open or close an isolated second-brain write session (worktree + PR). Use before writing knowledge when multiple agents share one knowledge remote.
- ager-validate: Validate AGER bundle structure and field rules, then delegate link/graph validation to okf-plugin (okf-graph-eng). Use before merge or after authoring multi-agent graphs.

Descriptions come from the frontmatter of each SKILL.md, punctuation lightly normalized.