Guide · Agent skill
How to use the Narrative Team Orchestrator agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The team-narrative skill is a reusable instruction set for Claude that makes the agent plan story content the way a small narrative team would: a narrative director, writer, world-builder, level designer and localization lead, each handling their part in a fixed sequence of phases. It is aimed at game developers and narrative designers who are writing dialogue, lore or story-driven levels and want the output checked against existing canon. After this guide you will know what the skill does, how to install it, how to trigger it with a request and when to choose a lighter tool instead.
Narrative Team OrchestratorDonchitos/Claude-Code-Game-Studios · MIT Download the skill Quick start
- 1Download and unzip. Download the ZIP from this page and unzip it into ~/.claude/skills/ or your project's .claude/skills/ folder.
- 2Confirm the repository files. Make sure .claude/hooks/yaml-helper.sh and the .claude/docs files are present in the project where you will run the skill.
- 3Gather your canon. Put your story bible or tone guide, character and faction notes and world rules where the agent can read them, such as under design/narrative/.
- 4Invoke the skill. Run /team-narrative with a specific description of the scene, lore area or narrative level.
- 5Choose a team size and mode. Pick individual, small or studio for team.size, and collaborative, guided or autonomous for automation, depending on how much control you want.
- 6Answer decision points. When the agent presents proposals through AskUserQuestion, choose the option that fits your canon and continue.
- 7Review the drafts. Read the draft files written under production/narrative/ and check them against your existing canon before using them.
What is an agent skill, and what does this one do?
An agent skill is a folder containing a SKILL.md file that tells Claude how to handle a certain kind of task. When the skill is invoked, Claude follows the process written in that file instead of improvising. The team-narrative skill is one of these, and it is licensed under MIT.
When you invoke it with a description of a story scene, lore area or narrative level, the skill tells the agent to read any existing story bible, tone guide, character and faction notes and world rules first. It then works through phases, playing or spawning specialist roles such as narrative director, writer, world-builder, level designer and localization lead. At decision points it asks you to choose between proposals.
- Reads existing canon before drafting, and is told not to invent canon if none exists.
- Runs a multi-step pipeline rather than returning a single answer.
- Asks you to pick between proposals at phase transitions.
- Writes draft documents under production/narrative/ for the content slug you are working on.
Install the team-narrative skill
Download the ZIP from this page. Unzip it into one of two places. For a skill you want available in all your projects, unzip it into ~/.claude/skills/. For a skill that should only apply to one project, unzip it into that project's .claude/skills/ folder.
You can also upload the skill in the Claude apps. Go to Settings, then Capabilities, then Skills, and upload it there.
- Personal install: ~/.claude/skills/ (available across projects).
- Project install: the .claude/skills/ folder inside your project.
- Claude apps: Settings → Capabilities → Skills.
Check what the skill needs before you run it
The skill is written for Claude Code and depends on parts of the Donchitos Claude-Code-Game-Studios repository. It needs the .claude/hooks/yaml-helper.sh script and the .claude/docs files to be present, along with Bash access so the agent can run the yaml-helper.sh config script. If those files are missing, the configuration step cannot resolve its settings.
The skill's allowed tools include file reading and searching, writing and editing, spawning subagents and task tracking. Its Bash permission is limited to the resolve_config call into yaml-helper.sh.
- Claude Code.
- The Donchitos Claude-Code-Game-Studios repository's .claude/hooks/yaml-helper.sh and .claude/docs files.
- Bash access to run yaml-helper.sh resolve_config.
Trigger the skill with a request
Invoke the skill with the /team-narrative command followed by a description of the content. The argument hint in the source is [narrative content description] with an optional --review setting of full, lean or solo. Without any argument the skill prints usage guidance and stops without spawning any agents.
Describe the scene, faction or area as specifically as you can. The source's own examples are boss encounter cutscene, faction intro dialogue and tutorial narrative. Keep your description to what the content is, so the agent knows which canon files to read.
- Good: faction intro dialogue for the river guild, before the first quest.
- Good: boss encounter cutscene, reveal the mill owner's betrayal.
- Too vague for this skill: write a line of dialogue.
/team-narrative [narrative content description] [--review full|lean|solo]Understand team size and how many agents run
The team.size setting decides which agents are active. The default is individual, where only the writer runs and the narrative director is brought in only when a pillar conflict is found. The small setting adds the narrative director, and adds the localization lead and world-builder when review_mode is full. The studio setting runs all six team agents regardless of review_mode.
Before spawning anything, the skill is meant to state in one line which agents will run and which will be consulted through another agent instead. At the default, most runs are writer-led, with other roles consulted through the writer.
- individual (default): writer only.
- small: narrative director and writer, plus localization lead and world-builder when review_mode is full.
- studio: all six team agents, as the full pipeline is documented.
Choose an automation mode and review depth
The automation setting controls how decisions move forward. In collaborative mode you must approve each phase before the next one starts. In guided mode the pipeline advances automatically unless a phase is blocked. In autonomous mode it runs end to end and records each phase outcome. Decisions in categories marked always-ask still prompt you in every mode.
The review_mode setting controls how deep director gates go. The source states that this pipeline has no director gate, but the ND-CONSISTENCY review in Phase 4 always runs regardless of review_mode. The --review argument overrides review_mode for a single run.
- collaborative: you approve every phase.
- guided: advances automatically unless blocked.
- autonomous: runs end to end and logs each phase.
- Configuration defaults are described in .claude/docs/config-resolution.md and automation modes in .claude/docs/automation-modes.md.
Safety notes: scripts, files and network
The skill runs a script. Its only permitted shell call is the resolve_config command in yaml-helper.sh, which reads configuration settings. The skill writes its drafts under production/narrative/[content-slug]/ and the source says the finished documents belong elsewhere, so review files before you move them into your project.
The source states that the skill does not use the network. Read the SKILL.md file and the helper script yourself before installing if your studio has review rules for third-party skills.
- Runs scripts: yes, the yaml-helper.sh config call only.
- Uses network: no.
- Writes draft files under production/narrative/[content-slug]/.
- Review all drafts before they enter canon.
What people use it for
- Drafting a boss encounter cutscene and checking it against the existing villain backstory.
- Writing a faction intro dialogue that matches the established politics of the region.
- Designing a tutorial level whose environment tells part of the story.
- Reviewing new lore entries for contradictions with the world rules.
- Preparing dialogue that a localization lead can check for non-localizable strings and cultural assumptions.
Frequently asked questions
Is the team-narrative skill free?
The skill is licensed under MIT, as stated on this page. The licence is the only price information the source gives, and you still need Claude Code and access to its features to run it.
What do I need to run the team-narrative skill?
You need Claude Code, the Donchitos Claude-Code-Game-Studios repository's .claude/hooks/yaml-helper.sh and .claude/docs files, and Bash access so the config script can run. The skill will not work well without those files.
Where do I install the team-narrative skill?
Unzip the downloaded ZIP into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. You can also upload it in the Claude apps under Settings, then Capabilities, then Skills.
Why does the skill ask me questions instead of just writing?
At each phase transition it presents proposals from the subagents as selectable options so you can choose the direction. How often it stops depends on the automation mode you have set, with collaborative asking at every phase.
Does team-narrative run scripts or use the internet?
It runs a script: its only permitted shell call is the resolve_config command in yaml-helper.sh. The source states it does not use the network.
When should I use something else instead?
For a quick one-off dialogue line the multi-phase pipeline and configuration step add overhead you do not need. If your project has no written story canon, the skill will not invent canon, so the results will only be as good as the material you provide.
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 team-narrative if you are a game developer or narrative designer planning dialogue, lore or story-driven levels and want each new piece checked against existing canon in a structured way. Skip it for quick one-off lines or for projects with no written story canon, where a simpler prompt will be faster and the output will be limited by what you can supply.
Download the skill