Skip to content
Get packs

Guide · Agent skill

How to use the Control Manifest Builder agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The create-control-manifest skill is an agent skill for Claude Code that reads a game project's accepted architecture decision records (ADRs) and writes a short rules sheet for programmers. The sheet lists what to do and what never to do, grouped by architectural layer, with each rule traced to the ADR it came from. This guide explains what an agent skill is, how to install this one, how to trigger it, and what to check before you approve the output.

Control Manifest BuilderDonchitos/Claude-Code-Game-Studios · MIT

Quick start

  1. 1Download and unzip the skill. Download the ZIP from the directory page and unzip it. Place the folder in ~/.claude/skills/ for all projects, or in the project's .claude/skills/ folder for one project.
  2. 2Check your project setup. Confirm the project uses the Claude Code Game Studios folder structure, has a project.yaml file, and has accepted ADRs in docs/architecture/ with Status sections.
  3. 3Confirm engine reference docs. Make sure docs/engine-reference/ contains the reference files for your chosen engine.
  4. 4Ask for the rules sheet. In Claude Code, request programmer rules or a cheat sheet from the ADRs, or ask for the manifest to be regenerated.
  5. 5Review the preview. Check the covered ADRs, excluded ADRs, rule counts and unresolved items. Answer whether the summary looks complete.
  6. 6Approve the write. Approve the write to receive docs/architecture/control-manifest.md. Regenerate it whenever an ADR changes.

What is an agent skill?

An agent skill is a folder containing a SKILL.md file. The file tells an AI agent when to use a capability and how to carry out a specific task. Once a skill is installed, the agent can load it automatically when a request matches the skill's description, or you can invoke it directly.

This skill is one of the Claude Code Game Studios workflow tools. It expects that folder layout and the architecture documents that workflow produces.

What the skill makes the agent do

The agent reads the project's accepted ADRs, the technical preferences file, and the engine reference docs. It then writes a flat rules sheet called a control manifest. The sheet separates rules into four layers: foundation, core, feature, and presentation.

The manifest says what programmers must do and must never do. It also records performance guardrails and forbidden APIs taken from the engine's deprecated-API list. The ADRs explain why a decision was made. The manifest records what to do about it.

  • Foundation: scene management, event architecture, save/load, engine start-up
  • Core: core gameplay loop, main player systems, physics and collision
  • Feature: secondary systems and mechanics, AI
  • Presentation: rendering, audio, UI, VFX, shaders

How to install the skill

Download the skill's ZIP file from the directory page and unzip it. To make it available across your projects, place the unzipped folder in ~/.claude/skills/. To make it available only in one project, place it in that project's .claude/skills/ folder.

In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

  • Personal install: ~/.claude/skills/
  • Project install: .claude/skills/ inside the project folder
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP

What you need before running it

The skill is built for a specific setup. Check these items before you start, or the skill will stop early.

It also needs an engine reference for the engine you use, under docs/engine-reference/, and a project.yaml config file.

  • Claude Code, with the Claude Code Game Studios folder structure (.claude/skills, .claude/scripts, .claude/docs)
  • A Bash shell, for the helper scripts yaml-helper.sh, adr-status.sh and md-section.sh
  • Existing docs/architecture/adr-*.md files, each with a Status section
  • A project.yaml config file
  • Engine reference docs under docs/engine-reference/ for your chosen engine

How to trigger the skill with natural requests

You do not need a special command. Ask in plain language for programmer rules, a cheat sheet, or a regenerated manifest. The skill's description covers requests such as these.

You can also pass the argument update to regenerate the manifest from the current ADRs, and --review with full, lean or solo to set the review mode for one run.

  • Make me a cheat sheet of programmer rules from our ADRs.
  • Create the control manifest from the accepted architecture decisions.
  • Regenerate the control manifest, an ADR has changed.
  • Update the control manifest from current ADRs.
/create-control-manifest update --review lean

Walk through a run

First the agent loads the inputs. It counts the ADRs by status before reading any of them. If there are no ADRs, it stops and asks you to run /architecture-decision. If none are Accepted, it asks whether to continue with Proposed ADRs or stop. It never writes an empty manifest silently.

Next it reads only four sections from each Accepted ADR: Decision, Alternatives Considered, Performance Implications and Engine Compatibility. It then adds global rules for naming, performance budgets, forbidden APIs and approved libraries.

Before writing anything, the agent shows a preview. It lists the ADRs covered, the ADRs left out with their status, the rule counts per layer, and any engine entries it could not resolve. You answer whether the summary looks complete. The file is written only after you say yes.

  • Stop condition: no ADRs found
  • Stop condition: ADRs have no Status sections, so acceptance cannot be determined
  • Preview: covered ADRs, excluded ADRs, rule counts, unresolved items
  • Write: docs/architecture/control-manifest.md, after your approval

Safety notes: scripts and network

The skill runs Bash helper scripts from the project's .claude/scripts and hooks folders. Its tool permissions limit Bash use to the specific scripts it needs. The skill is documented as not using the network.

The only file it writes is docs/architecture/control-manifest.md, and only after you approve the preview. Read the SKILL.md file before installing if you want to check the scripts it calls.

  • Runs scripts: yes (yaml-helper.sh, adr-status.sh, md-section.sh)
  • Uses network: no
  • Writes: docs/architecture/control-manifest.md, after your yes

Tips for better results

Run the skill after an ADR is accepted or revised, so the manifest stays in step with the decisions. Keep ADR Status sections current, because the skill decides what is Accepted from them.

Review the preview's list of excluded ADRs. A rule from a Proposed or Deprecated ADR should not reach programmers by accident.

Check the unresolved list. Engine entries that only apply to 2D or 3D projects may be scoped or left out, so decide whether they belong.

  • Regenerate after every ADR acceptance or revision
  • Keep Status sections accurate in each adr-*.md file
  • Read the excluded and unresolved lists before approving

What people use it for

  • A technical lead produces a one-page cheat sheet of must-do and never-do rules for programmers, grouped by layer.
  • A solo developer regenerates the manifest after accepting a revised ADR for the physics system.
  • A studio checks that the forbidden-API list from the engine's deprecated-API file reaches the programmers' rules.

Frequently asked questions

Is the create-control-manifest skill free?

The skill is published under the MIT licence. The source material here does not cover pricing for Claude Code or the Claude apps, so check those products' own terms for any costs.

What do I need to run the create-control-manifest skill?

You need Claude Code with the Claude Code Game Studios folder structure, a Bash shell for the helper scripts, accepted ADR files with Status sections, a project.yaml config, and engine reference docs under docs/engine-reference/. Without ADRs the skill stops and asks you to create them first.

Why does the skill stop with a message about Status sections?

Each ADR must have a Status section so the skill can tell whether it is Accepted. If none has one, acceptance cannot be determined, and the skill stops. Run the architecture decision retrofit on each file, then try again.

Can I use this skill for a non-game or non-engine project?

The source says the skill is not for non-game or non-engine projects, because it expects engine reference files and the studio's specific folder layout. Use it only in a project that matches that layout.

How do I update the manifest after changing an ADR?

Run the skill again with a request to regenerate the manifest, or pass the update argument. It reads the current ADRs and shows a preview, then replaces docs/architecture/control-manifest.md after you approve.

Are there alternatives to this skill?

The source does not compare this skill with other tools. You can write the programmer rules by hand from your ADRs, or use another agent skill that suits your workflow. Check the directory listing for related skills.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you work in the Claude Code Game Studios workflow, already have accepted ADRs, and want a rules sheet for programmers that stays tied to those decisions. Pick something else if your project has no ADRs, is not a game or engine project, or does not follow that folder layout.

Download the skill