LLM Easy Shell Lite

Read-only subset of LLM Easy Shell →. 9 safe exploration commands, no writes, no PIE control, no Live Coding. The minimum you need to look at a running UE editor without risk.

Skill file: skills/llm-easy-shell-lite/SKILL.md

When the Agent Loads It

The agent loads this skill when the user asks to:

List actors in the current level
Inspect an actor / component / asset (read-only)
Search for actors or assets by name pattern
Discover an actor's available functions and properties (before calling)
Check editor state (info, pieState, livecoding)
Read editor error log or Niagara validation messages
Query PIE runtime state (gamestate)

Triggers: ls, cat, find, discover, info, help, gamestate, log, msglog, /Actor/, /Assets/, /Game/, /Level/.

Does NOT trigger for: any write operation (set / spawn / rm / cp / mv / call / save / new / addcomp / play / stop / pause / undo / redo / reload), screenshots, C++ coding, AngelScript scripting.

For any of the above, the agent should load the full LLM Easy Shell → skill.

When to Use Lite vs Full

Scenario	Use
"What actors are in this level?"	Lite ✅
"What does this actor's material look like in PIE?"	Lite + screenshot path on Full (Lite has no screenshot)
"Move this light to the corner"	Full
"Discover the function signature before I decide whether to call it"	Lite ✅
"I want a safe read-only sandbox"	Lite ✅
"I need the full Python sub-command suite"	Full

Rule of thumb: default to Lite for exploration; switch to Full when the user requests a change.

Prerequisites

Requirement	Notes
UE 5.7+	Engine version
LLM Easy Shell Lite plugin	Install from Fab or copy from this repo to `Plugins/`
Editor must be running	Same as Full — commands RPC to a live editor process
`llm-shell` CLI	Build artifact: `skills/llm-easy-shell-lite/llm-shell/llm-shell.exe` (Windows) / `llm-shell` (Linux/Mac)
LLMEasyShellLite plugin enabled	Edit → Plugins → AI → LLM Easy Shell Lite → restart editor

The 9 Read-Only Commands

bash

LITE=skills/llm-easy-shell-lite/llm-shell/llm-shell

# Browsing
$LITE -c "ls /Level/Actors"                    # List actors
$LITE -c "ls /Assets"                          # List asset directories
$LITE -c "cat /Actor/MyPawn"                   # Inspect an actor
$LITE -c "cat /Actor/MyPawn/MyComponent"       # Inspect a component

# Search
$LITE -c "find Actor BP_Liquid*"               # Search actors by name
$LITE -c "find Asset *Liquid*"                 # Search assets by name
$LITE -c "find /Game *Glass*"                  # Search by path glob
$LITE -c "find /Assets *Glass*"                # Same, /Assets alias

# Discover (call these BEFORE you switch to Full to set/call)
$LITE -c "discover /Actor/MyPawn --funcs"      # Available functions
$LITE -c "discover /Actor/MyPawn --props"      # Available properties

# Status
$LITE -c "info"                                # Project + editor + PIE state
$LITE -c "help"                                # List all commands
$LITE -c "gamestate"                           # PIE wave / health / gold / enemies
$LITE -c "log --errors"                        # Editor error log
$LITE -c "msglog --niagara"                    # Niagara validation messages

Convention: the recommended workflow is — discover with Lite, then act with Full. Lite gives the agent a "dry run" view that prevents typos in the eventual set / call.

Path Conventions

Path	Meaning
`/Actor/{label}`	An actor in the current level
`/Actor/{label}/{comp}`	A component on that actor (alias-matched)
`/Game/{path}`	An asset under `/Game/`
`/Assets/{path}`	Alias of `/Game/{path}`
`/Class/{name}`	A UClass

Plugin Abilities (Limited Subset)

The four LLM series plugins also register read-only sub-commands that Lite can see via help:

Ability	Source plugin	Read-only sub-commands
`llmstatetree`	LLMStateTree	`inspect`, `list-types`, `describe-type`, `export-schema`, `export`
`ui`	LLMDynamicUI	`inspect`, `list-types`, `export-schema`, `export`
`material`	LLMMaterial	`inspect`, `list-types`, `describe-type`, `export-schema`, `export`

These are safe to use with Lite. Sub-commands that write (generate, copy, delete, compile) technically work because Lite shares the handler instance with those plugins — use them with caution and accept the risk that they violate the read-only contract.

What Lite Does NOT Have

❌ Any write operation (set / spawn / rm / cp / mv / call / save / new / addcomp)
❌ Any PIE control (play / stop / pause)
❌ livecoding / restart compile cycle
❌ log --compile (compile log capture — needs Full or an Advanced skill)
❌ Screenshots (PIE path or Win32 path)
❌ -q / -j CLI flags (Lite does not implement them — read stdout directly)
❌ python <name> 25+ sub-commands

For any of these, switch to LLM Easy Shell →.

Port

Lite uses TCP port range 15201–15250 (50 ports). The default is 15201.

Full uses 15151–15200. The two ranges never overlap — you can run Lite and Full side by side on the same machine without coordinating ports.

Port file: llm-shell/.current_port.

Troubleshooting

Problem	Solution
`Connection refused`	UE editor is not running — start it
`Command not found: <X>`	`<X>` is a write command Lite does not implement. Use Full.
Output is noisier than expected	Lite does not implement `-q` / `-j`. Pipe through `jq` or read stdout directly.
A plugin's write sub-command worked anyway	Lite shares the handler with the plugin; Lite's "read-only" is enforced at the skill's command list, not at the C++ level. Treat it as advisory.

llm-easy-shell → — full read/write version.
All four asset-authoring skills (dynamic-ui, material, statetree, metasound) — use Lite to inspect existing assets, then the asset-authoring skill to author new ones.

LLM Easy Shell Lite ​

When the Agent Loads It ​

When to Use Lite vs Full ​

Prerequisites ​

The 9 Read-Only Commands ​

Path Conventions ​

Plugin Abilities (Limited Subset) ​

What Lite Does NOT Have ​

Port ​

Troubleshooting ​

Related Skills ​