In This Tutorial
1 Why Curation?
CLAUDIUS ships ~230 commands across 26 categories. If every command's full documentation hit Claude's context on every prompt, you'd burn 12,000+ tokens before saying a word. Worse, the AI would have to scan irrelevant commands every time it answered a question.
v3's answer is to track every command in a registry and let you mark each one enabled or disabled per project. The auto-generated Claudius/CLAUDIUS.md only spells out enabled commands in detail; disabled ones appear in a one-line index entry so the AI can still see they exist and offer to enable them.
Curation prioritizes signal density over completeness. Your AI sees less but knows what's possible. When you ask for something it can't currently do, the response includes the exact enable call to surface it.
2 The Command Registry
Every command has four pieces of metadata in the registry:
- Category — e.g.
level,blueprint,sequencer. - Description — one-line summary that lands in
CLAUDIUS.md. - Tags — comma-separated workflow tags like
core,fps,animation,shot. - Default enabled — whether the command is on out of the box.
The same registry tracks user Python commands and skill packs. There's no second-class status: a Python script with # @claudius headers shows up next to a built-in C++ command in the index.
Inspect the live registry:
# Category summary with enabled counts { "category": "help", "command": "list_categories", "parameters": {} } # Fuzzy-search the full index { "category": "help", "command": "find", "parameters": { "query": "render mrq" } } # Just the disabled-but-available ones { "category": "help", "command": "list_disabled", "parameters": {} }
3 Workflow Presets
A preset is a bundle of enable rules. Loading fps turns on every command tagged core, fps, blueprint, level, animation, debug — about 105 commands.
Available presets:
minimal # core meta-commands only — smallest viable surface fps # first-person shooter scaffolding platformer # 3D platformer / movement-heavy level_design # whitebox layout, set dressing animation # generic animation asset authoring animation_studio # shot-based commercial work (NEW) open_world # open-world, PCG, large scenes everything # enable every command in the registry
Switch preset at any time:
{
"category": "config",
"command": "load_preset",
"parameters": { "preset": "animation_studio" }
}
Loading a preset is non-destructive in the sense that nothing is deleted — it just rewrites the enable set. Your skill packs, user commands, and profile stay intact. Claudius/CLAUDIUS.md regenerates immediately.
Load a preset as a baseline, then turn individual commands on or off with config.enable_command / config.disable_command. Your tweaks persist across sessions in Saved/Claudius/config.json.
4 The Ask-Before-Enable Policy
When the AI tries to call a disabled command, the plugin's response depends on your auto_enable_policy:
ask # default. Returns command_disabled with the suggested enable call. # The AI surfaces the suggestion and waits for user confirmation. always # Plugin silently runs config.enable_command and processes the call. # Trade-off: faster iteration, less visibility into what's getting added. never # Plugin returns a hard failure. User must enable manually via UI/JSON. # Use when you want strict control over what the AI can reach.
Sample command_disabled response under the default ask policy:
{
"success": false,
"error_code": "command_disabled",
"message": "sequencer.create_sequence is currently disabled",
"suggested_call": {
"category": "config",
"command": "enable_command",
"parameters": { "command": "sequencer.create_sequence" }
}
}
Change the policy at any time:
{
"category": "config",
"command": "set_auto_enable_policy",
"parameters": { "policy": "ask" }
}
5 Discovery Commands
The help.* and config.* categories are always enabled, regardless of preset. They're the discovery primitives the AI leans on:
help.ping # round-trip check help.find # fuzzy-search the full index help.list_categories # category + enabled count summary help.list_disabled # every disabled-but-available command help.list_packs # installed skill packs config.status # profile + health + enabled counts + paths config.list_presets # preset names
The config.status response is the canonical "what's the state" probe — profile, enabled command count, paths, link state, all in one call:
{
"output": {
"profile": { "workflow": "animation_studio", ... },
"setup_complete": true,
"auto_enable_policy": "ask",
"enabled_commands": 112,
"total_commands": 230,
"installed_packs": 2,
"claudius_md_path": ".../Claudius/CLAUDIUS.md",
"user_claude_md_path": ".../CLAUDE.md",
"user_claude_md_state": "linked",
"estimated_tokens": 8120
}
}
6 Common Patterns
Enable a single command for one task
"I just want to render this one sequence." Enable, do the work, optionally disable again to keep the surface clean.
{ "category": "config", "command": "enable_command",
"parameters": { "command": "sequencer.render_sequence" } }
Enable an entire category
For sustained work in one area — e.g. you're doing a sequencer-heavy day:
{ "category": "config", "command": "enable_category",
"parameters": { "category": "sequencer" } }
Save and share a profile
Export the current profile + enable set as JSON, hand it to a teammate:
{ "category": "config", "command": "export_profile", "parameters": {} }
{ "category": "config", "command": "import_profile",
"parameters": { "profile_json": "{...}" } }
"What can CLAUDIUS do for me right now?"
One call gives you a category-by-category snapshot:
{
"output": {
"categories": [
{ "name": "level", "total": 28, "enabled": 22 },
{ "name": "sequencer", "total": 13, "enabled": 13 },
{ "name": "control_rig", "total": 7, "enabled": 2 },
...
]
}
}