PoBruno

Use this skill whenever the user's request touches the UE5 project this skill is installed in — analyzing a bug, observing editor/scene/runtime state, reading or editing Blueprints/materials/animation/widgets/levels/sequencer/MRQ, spawning or modifying actors, running PIE, taking screenshots, or anything that would otherwise require manually opening the Unreal Editor. The `unreal-agent` MCP server gives you live, structured, editor-accurate access to the project — prefer it over guessing from source files alone.

Step-by-step playbook for installing mcp-unreal-agent into an end user's UE5 project. Read when the user asks to install the MCP, set up the agent, or follow the install prompt.

How the C++ context bridge works — when a Blueprint references a C++ class / function / variable, how the agent locates and reads the source. Read when implementing or debugging tools that surface C++ source to the agent.

How to design and register a new MCP tool — input schema (Zod), output contract, descriptions, tests. Read before adding any new tool to Tools/src/tools/.

Designing composite UE5 tools and chaining existing ones via the ID-passing contract. Use when adding a new tool, designing a multi-step flow, or refactoring primitives into a composite atomic flow.

UE5 editor API cheat sheet for AI agents — Subsystem map, deprecated→modern translations, transaction patterns, where Python reaches and where C++ is required. Read when picking the right API for a tool implementation or when porting deprecated code.

Passive context — you control a live Blender session through the blender-agent MCP. Load this whenever the user asks to model, rig, animate, texture, light, render, or export anything in 3D, or mentions Blender. It tells you the operating loop, the tool map, and the gotchas that bite first-timers.

Blender Python API cheat sheet for AI agents — bpy.data vs bpy.ops, the main-thread rule, context overrides, mode switches, where to reach for bmesh vs operators. Read when picking the right API for a tool implementation.

Step-by-step playbook for installing mcp-blender-agent into an end user's agent harness (Claude Code / Copilot / Cursor / Claude Desktop). Read when the user asks to install the MCP or follow the install prompt.

How to design and register a new MCP tool — input schema (Zod), output contract, descriptions, tests. Read before adding any new tool to Tools/src/tools/.

Designing composite Blender tools and chaining existing ones via the ID-passing contract. Read when adding a new tool, designing a multi-step flow, or refactoring primitives into a composite atomic flow.

Use ONLY when a phase needs a tool that is not yet wired up — most often a headless browser to capture a screenshot for visual/perceptual verification of a UI. Synthesizes the MCP server on demand (in an isolated container), registers it in opencode.json, validates it, and notes credential placeholders. Not for MCPs already present in the manifest.

Use when creating, authoring, or optimizing an opencode skill (a SKILL.md procedure) — including when a phase keeps re-deriving the same how-to and you want to capture it. Guides an evaluation-driven workflow: interview, draft, generate trigger tests, install only the validated version.