# Skills, playbooks and graph orchestration | Free Resource

> A free six-slide model of how skills, playbooks, agents and graph orchestration fit together, with a node contract for bounded, verifiable AI work. Read it here or download the PDF.

Source: https://andrewheldy.dev/resources/ascensionos-skills-playbooks-graph

---

Free resource · Framework

## Skills, playbooks and graph orchestration

A six-slide model for building AI work out of reusable parts: skills become playbooks, playbooks equip specialist agents, and a graph routes each task through the smallest context that can do it.

[Download the PDF](https://andrewheldy.dev/resources/ascensionos-skills-playbooks-graph.pdf) 6 slides, PDF · 40 KB · free, no signup

A conceptual model, not a product tour. The diagrams are illustrative examples of how the parts fit together.

Overview

### System architecture

Reusable competencies become playbooks. Playbooks equip specialist agents. Graph orchestration routes work through the smallest necessary context.

1.  Tools

    Capability

2.  Skills

    Competence

3.  Playbooks

    Procedure

4.  Agents

    Specialization

5.  AscensionOS

    Orchestration


Tools → Skills → Playbooks → Agents → AscensionOS

Tools, skills, playbooks and agents stack up into AscensionOS, each layer built from the one before. Illustrative animation.

Capability becomes reliable execution through explicit composition.

01 · The stack

### One canonical stack. No duplicated intelligence.

Playbooks reference canonical skills. Agents select them; they do not carry private copies.

1.  01

    #### Tools

    GitHub · terminal · browser · APIs

2.  02

    #### Skills

    Reusable procedures with verification

3.  03

    #### Playbooks

    Sequences · branches · bounded loops

4.  04

    #### Agents

    Specialists with selected capabilities

5.  05

    #### Agent OS

    Policy · routing · memory · permissions

6.  06

    #### AscensionOS

    Classify · route · state · reconcile


02 · Skills + playbook

### Skills are atoms. Playbooks are compositions.

Canonical skills

1.  Skill

    inspect-repo

    canonical

2.  Skill

    research-web

    canonical

3.  Skill

    write-plan

    canonical

4.  Skill

    implement

    canonical

5.  Skill

    run-tests

    canonical

6.  Skill

    verify

    canonical


Example playbook: ship feature

1.  Skill

    inspect-repo

2.  Skill

    write-plan

3.  -   Skill

        security-audit

    -   Skill

        implement

    -   Skill

        docs-update


4.  Skill

    integration-test

5.  Gate

    verify


↺ failure → targeted repair

Run itStepResetSimulate a failure

Press Run to walk this row one step at a time.

A failed check sends work back for a targeted repair, not a restart. The budget is one pass plus two targeted repairs. Illustrative animation.

03 · Agent OS

### An agent is a configured worker, not a giant prompt.

The agent OS determines what the specialist can do, how it works, what context it sees, and how success is proven.

Specialist

Software Engineer

implementation quality

Skills

architecture · coding · testing

Tools

GitHub · terminal · filesystem

Playbooks

ship-feature · fix-bug · refactor

Model policy

cheap explore · reason · verify

Policies

inspect first · smallest diff · proof

04 · Graph-orchestrated prompt

### How a prompt becomes executed work.

User: “Build authentication for the dashboard, preserve the existing design system, test it, and document the change.”

Route

1.  Input

    Prompt

    goal + constraints

2.  Orchestrator

    Classify

    intent · risk · scope

3.  Router

    Select agent

    Engineer OS

4.  Playbook

    ship-feature

    graph loaded

5.  State

    Checkpoint

    canonical state


Run itStepReset

Press Run to walk this row one step at a time.

Execute

1.  Skill

    inspect-repo

    GitHub

2.  Skill

    write-plan

    state

3.  Skill

    implement

    filesystem

4.  Skill

    run-tests

    terminal

5.  Gate

    verify

    evidence


Run itStepReset

Press Run to walk this row one step at a time.

Each node loads only its own skill instructions, state and tools; the rest of the project stays out of working context. Illustrative animation, not measured data.

Context policy

Load only the current node’s skill instructions, relevant project state and permitted tools. Keep everything else out of working context.

Result: explicit ownership + bounded retries + mechanical verification.

05 · Node contract

### Every graph node gets a small contract.

This is the layer that turns graph orchestration from prompt spaghetti into production infrastructure.

Objective

Implement dashboard authentication.

Input state

Approved plan · architecture constraints.

Skills

implement-code · secure-auth-patterns.

Tools

filesystem · terminal · package manager.

Model routing

Coding model; escalate on architecture ambiguity.

Budget

1 pass + 2 targeted repairs.

Acceptance

Auth tests + regression tests + no secrets.

Output

Patch · evidence · changed files · risks.

North star: minimum sufficient context, capability, orchestration and verification.

### Want to see where this fits in your business?

A complimentary conversation decides whether an Audit, a build, training, or nothing is the right next step.

[Discuss your business→](https://andrewheldy.dev/start)[All resources](https://andrewheldy.dev/resources)