User Guide

Current — this is the living version. Historical snapshots are linked from Index-Releases.

User Guide

Day-to-day use of okf-graph-eng on an OKF repository.

Concepts

The plugin treats your OKF bundle as a dual graph:

1. Knowledge graph — datasets, metrics, runbooks, APIs, references
2. Agent / harness graph — AgentNode, Workflow, SharedState, DecisionRecord, ToolCapability, TicketLink

Edges are absolute Markdown links ([Label](/path/to/concept.md)). Optional typed relations live in frontmatter:

links:
  - target: /agents/writer.md
    rel: routes_to

Common workflows

Scaffold a bundle

• Slash: /okf-init
• Or ask to initialize an OKF graph-engineering bundle (skill: okf-init-graph)

Default tree: .okf/ with agents/, workflows/, knowledge/, decisions/, shared/, tickets/, plus index.md and log.md.

Author concepts

Use skill okf-author (or /okf-author). Every concept needs at least type, title, description, timestamp.

Impact before structure changes

python3 scripts/okf-graph.py impact <bundle> <concept>
# or /okf-impact

Prefer this before renaming or splitting high-degree agents, workflows, or shared state.

Progressive disclosure packs

Default: 2 hops, ~20 nodes.

python3 scripts/okf-graph.py pack <bundle> <concept> --hops 2 --max-nodes 20

Validate

python3 scripts/okf-graph.py validate <bundle>
# or /okf-validate

Tickets (WikiTicket / worklog)

This repo is managed with WikiTicket SDD. Map work items into OKF:

bin/worklog fold | python3 scripts/okf-ticket-link.py emit --bundle sample-okf --open-only

Sample bundle

sample-okf/ models this plugin as both knowledge and agent graph. Use it as a template and for demos.

See also

• CLI-Reference — scripts and flags
• Plugin-Guide — install on Claude Code / Grok Build
• Roadmap — generated from the worklog