| name | improve-environment |
| description | Audit and improve track environment visual quality — lighting, materials, terrain, post-processing, probes, LODs. Dispatches unity-environment-artist agent. |
Environment Quality
Evaluate and improve how a track scene looks by auditing against a 25-item quality checklist.
Arguments
scene (optional): Track or scene name (e.g., "Outpost", "PhysicsTest"). Defaults to currently loaded scene.
mode (optional): audit (default) or improve. Audit is read-only; improve implements changes.
focus (optional): Narrow scope to a specific category — lighting, materials, terrain, postprocess, probes, composition. Default: all categories.
Workflow
Step 1: Determine Target Scene
If scene argument provided:
- Look for
Assets/Scenes/{scene}Track.unity or Assets/Scenes/{scene}.unity
- Read the track's folder for EnvironmentSettings:
Assets/Tracks/{scene}/
If no argument:
- Query the currently active scene via MCP
Step 2: Pre-Read Context
Read these files ONCE and extract relevant content for the agent prompt:
- Root CLAUDE.md — project conventions, builders, gotchas
.ai/knowledge/architecture/environment-quality.md — checklist, pillars, design principles
.ai/knowledge/architecture/terrain-workflow.md — terrain-specific guidance (if focus includes terrain)
.ai/knowledge/architecture/urp-rendering.md — renderer setup, volume profiles, shader rules
- Track-specific EnvironmentSettings SO (if it exists)
Step 3: Dispatch Agent — AUDIT Mode
Dispatch unity-environment-artist with full context:
Agent: unity-environment-artist
Mode: AUDIT
Scene: {target scene}
Focus: {focus category or "all"}
{paste pre-read context}
Query the scene hierarchy, lighting, volumes, probes, terrain, materials, and camera settings.
Score each of the 25 checklist items as PASS / WARN / FAIL.
Return a prioritized improvement list ranked by visual impact with effort estimates.
Step 4: Present Findings
Summarize the audit as:
## Environment Audit — {SceneName}
X PASS | Y WARN | Z FAIL
### Top Improvements (by visual impact)
1. [Category] Item — recommendation (Effort: Low/Med/High)
2. ...
Step 5: Offer Improvement
Ask: "Would you like me to implement any of these improvements?"
If the user selects items, proceed to Step 6. If not, stop here.
Step 6: Dispatch Agent — IMPROVE Mode
Re-dispatch unity-environment-artist with selected improvements:
Agent: unity-environment-artist
Mode: IMPROVE
Scene: {target scene}
Improvements to implement:
{numbered list of selected items}
{paste pre-read context}
For each improvement:
1. Determine if it maps to a builder parameter (use builder pipeline) or needs direct MCP edit
2. Implement the change
3. Save immediately after each modification
4. Commit modified files immediately
5. Report before/after values
Step 7: Verify & Report
After the agent completes:
- Check
mcp__UnityMCP__read_console for errors
- Verify commits via
git log
- Check
git status for orphaned .meta files
- Summarize changes made with before/after values
Triggers
This skill is appropriate when the user says things like:
- "make this look better"
- "improve the environment"
- "visual quality"
- "track looks flat/boring"
- "how's the lighting"
- "audit the scene"
- "what can we do to make it more realistic"