# Graph Flow Profile

> Author, validate, explain, dry-run, and inspect optional graph.flow.aiwg.io GraphPlaybooks over the AIWG Flow runtime substrate

- Skill: `jmagly/graph-flow-profile` (Agent Skill)
- Install (CLI): `npx skillmds@latest add jmagly/graph-flow-profile`
- Raw SKILL.md: https://api.skillmd.com/api/skills/jmagly/graph-flow-profile/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: jmagly (https://skillmd.com/u/jmagly)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/jmagly/graph-flow-profile

---


# Graph Flow Profile

Use this skill when a workflow requires conditional routes, guarded feedback,
explicit fan-in reduction, checkpoint/replay lineage, or a mixed-runtime graph
view. Keep ordinary DAGs, local loops, task trees, provider-native workflows,
and durable code in their native patterns when graph-wide state and routing do
not add operator value.

1. Start from a fixture in `agentic/code/addons/graph-pattern/fixtures/`.
2. Keep `apiVersion: graph.flow.aiwg.io/v1` and `kind: GraphPlaybook`.
3. Narrow each node's capabilities and permissions; never inherit authority
   from an upstream node.
4. Give cycles a finite guard, fan-in nodes an explicit join/reducer,
   side-effecting retries an idempotency key, and HITL nodes complete
   approve/deny/timeout routes.
5. Run `aiwg graph validate <file> --format json`, then
   `aiwg graph explain <file> --format json` before runtime dispatch.

The validator reuses the Composition Engine's Flow validation. A valid graph
profile is therefore a valid Flow projection; the graph profile is not a
standalone runtime.