hashgraph-online/awesome-codex-plugins / plugins/AUN-PN/unity-agent-workflows/skills/unity-agent-workflows
unity-agent-workflows
VerifiedUse for AI-assisted Unity work that needs live repo discovery, project-derived routing, runtime-owner proof, runtime-visible output hard stops, runtime numeric proof for repeated visible-output failures, state-step guards, multi-agent scope ownership, modular C#/asmdef safety, UI/scene/visual asset gates, data-first content changes, validation, cleanup proof, or durable workflow rules. Best when agents must prove the actual folder/module/scene/prefab/runtime owner before editing, especially runtime UI, generated assets, code graphs, tutorial/state flows, guided selection/action flows, overlay/dim source-bound mistakes, coordinate conversions, focus/highlight/marker/HUD alignment, predictive lead aiming, ballistic trajectories, spatial IoU verification, or repeated "fix still not visible" failures.
references/workflow-recipes.md
View on GitHub# Workflow Recipes
Read this only when a task needs a named recipe beyond `references/ai-workflows.md`.
## WF-0 Orientation
Use when placement or ownership is unclear.
1. Search exact screen text, method names, Unity object names, GUIDs, and runtime/content terms with `rg`.
2. Read likely owners and callsites.
3. Read graph output if present.
4. Fill the structure map from `references/project-structure-discovery.md` when placement matters.
5. Fill the Routing Card.
6. Ask only if two live owners are equally plausible and the choice changes user-visible behavior.
## WF-1 Narrow Bug Fix
1. Locate exact call path.
2. Patch the true owner.
3. Avoid architecture migration unless the boundary caused the bug.
4. Validate the smallest surface.
## WF-2 New Runtime Or Content Feature
1. Decide data-first vs code-first.
2. Prefer definitions/config for tuning and identity.
3. Put code-first feature behavior in the repo's owning feature/module/runtime path.
4. Route cross-module facts through contracts/events/bridges.
5. Extract a collaborator if a controller would gain a new responsibility group.
6. For tutorials, objectives, unlocks, guided selection/action steps, navigation gates, or other state machines, run the Runtime State Step Guard from `references/content-and-systems.md`.
## WF-3 Cross-Module Communication
1. Treat sibling feature imports as blocked.
2. Find the repo's existing gateway/event/contract boundary.
3. If none exists, create the smallest implementation-free contract.
4. Register implementation from the owner.
5. Update asmdefs/docs only when dependency structure changes.
## WF-4 Hub Deflation
1. Name the responsibility group being added or extracted.
2. Check line count and graph status when available.
3. Extract one focused collaborator with one reason to change.
4. Keep MonoBehaviour serialized fields stable if scenes depend on them.
5. Prove changed callsites, moved state, or removed coupling.
## WF-5 Data, Balance, Or Content
1. Search existing definitions/config/assets before editing code.
2. Prefer content data over branching inside runtime controllers.
3. Add C# only when the data declares behavior not currently handled.
4. Validate asset/catalog lookup names and runtime load paths.
5. Report exact constants/assets changed.
## WF-6 UI Or Screenshot Fix
1. Identify visible layer and runtime owner.
2. If the task involves any runtime visible output, focus, highlight, selection, click/tap target, visual target, bounds-type choice, marker, blocker, mask, world-to-UI label, HUD marker, modal dimming, duplicate names, hardcoded layout/position, or "do not guess", run the Runtime Visible Output Hard Stop from `references/runtime-owner-proof.md` before editing.
3. Separate layout/anchoring from readability/polish.
4. Change only the shown layer unless owner proof requires a direct dependency.
5. Preserve camera, background, composition, and runtime layout unless requested.
6. For world/local/screen/canvas/camera coordinate mismatch, read `references/coordinate-space-conversion.md`.
7. If an object is found but no marker exists, read `references/runtime-visible-targets.md` and report candidate rects/ambiguity before choosing.
8. Validate with screenshot, hierarchy, or serialized/runtime proof.
## WF-6R Repeated Visible Mismatch
Use when a UI/HUD/focus/highlight/marker/camera/world-to-UI fix already failed, looks unchanged, or lands in the wrong place.
1. Stop patching the same coordinate/layout/fallback owner.
2. Treat the user's latest screenshot or visible target wording as the current scope.
3. Read `references/runtime-visible-targets.md`, `references/coordinate-space-conversion.md`, and `references/unity-validation.md`.
4. Collect runtime numeric proof: source bounds, destination bounds/root, converted rect, final drawn rect, and runtime writer.
5. If values are missing, return a runtime probe plan only.
6. Let a checker fail the work if the next patch lacks numeric proof.
## WF-12 Multi-Agent Visible Or State Work
Use when a main agent coordinates worker agents for Unity visible-output, tutorial/state, guided selection/action, navigation, or runtime UI work.
1. Run the Sub-Agent Decision And Permission Gate from `references/ai-workflows.md`; do not spawn until the user approves in the same turn, unless the user already explicitly requested sub-agents.
2. Main agent owns scope before workers patch: load required references, fill the Routing Card, name the visible target or state flow, and list `Files allowed to touch` plus `Files explicitly not touched`.
3. Main agent assigns disjoint ownership: each worker gets a separate file set, owner chain, runtime target, state step, or validation slice. No overlapping writes unless main agent explicitly merges ownership first.
4. Workers stay inside the assigned write set. If a worker discovers a new owner, missing reference, or broader dependency, it reports back instead of patching outside scope.
5. Visible-output workers must provide runtime numeric proof when required: source bounds, destination bounds/root, converted rect, final drawn rect, and runtime writer.
6. State-flow workers must provide state-step proof: shown, clicked, opened, selected, applied, completed, persisted, plus old-save/default/reset paths when relevant.
7. Checker fails the run if the Routing Card is missing, write ownership overlaps, runtime numeric proof is missing for visible-output work, or state-step proof treats screen open/click/analytics as domain completion.
8. Main agent does final integration, validation, and closeout; workers do not broaden public rules, mirrors, package metadata, or repo instructions unless those files are in the allowed write set.
## WF-7 Visual Source Asset
1. Stop code editing.
2. Define the source asset being created/replaced/redesigned.
3. Generate or obtain the approved source asset.
4. Inspect transparency, import scale, orientation, 9-slice readiness, and pixel crispness.
5. Re-run architecture routing before integration code.
## WF-8 Compile Or Validation Repair
1. Capture exact error and assembly context.
2. Fix the smallest compile/runtime owner.
3. Avoid broad refactors unless the error is a boundary violation.
4. Re-run the same validation.
## WF-9 Architecture Decision
1. Keep always-read repo instructions short.
2. Put detailed process in a workflow doc or skill.
3. Put dependency facts in machine-readable rules when possible.
4. Put current runtime maps in architecture docs.
5. Create an ADR for significant decisions likely to be revisited.
6. Validate docs with `git diff --check`.
## WF-10 Cleanup Or Deletion
1. Prove unused status through code refs, YAML GUID refs, resource paths, asset refs, and runtime reachability.
2. Do not treat isolated graph nodes as deletion proof.
3. Separate generated/cache cleanup from source refactors.
4. Report what was not deleted and why.
## WF-11 Session History Or Rule Mining
1. Parse session files structurally.
2. Filter injected instructions, environment dumps, plugin text, and base64.
3. Group repeated patterns before editing rules.
4. Patch only the owning artifact.
5. Validate the rule/skill/docs change.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |