git:20260930.555d843 to git:20261001.26e74f3

27 added, 41 removed. Audit A to A.

- # Agent Instructions — `pascalorg/editor`

+ # Agents — `pascalorg/editor`

- Public, open-source home of `@pascal-app/{core,viewer,editor,mcp}` and the standalone editor app. Consumed both as npm packages and (in `pascalorg/private-editor`) as a git submodule.

+ Public, open-source home of `@pascal-app/{core,viewer,editor,nodes,mcp,cli,ifc-converter}` and the standalone editor app. Consumed as npm packages and, in `pascalorg/private-editor`, as a git submodule.

- ## Repo Shape

+ **Read `DECISIONS.md` first**: what Pascal is, the architecture invariants (E-IDs), the testing rule (T-001), and what is explicitly delegated to you. Everything else here is a map.

- | Path | Purpose |

+ ## Packages

+ | Path | Owns |

|---|---|

- | `packages/core` | Scene graph, node schemas, stores, event bus, core systems — pure logic, no Three.js. `src/capture/` holds the capture-session contracts published as `@pascal-app/core/capture` |

- | `packages/viewer` | Standalone 3D canvas: renderers, viewer systems, presentation state. `src/capture/` holds the capture runtime and reference layers published as `@pascal-app/viewer/capture` |

- | `packages/editor` | Editor UI components reused by the standalone app and embedders |

- | `packages/mcp` | MCP server and scene storage adapters |

- | `apps/editor` | Standalone editor app — composes `viewer` + `editor` + tools |

+ | `packages/core` | Scene graph, schemas, stores, commands, registry, services — pure logic, no runtime Three.js (E-001). `src/capture/` = capture-session contracts |

+ | `packages/viewer` | The 3D canvas: renderers, viewer systems, presentation state. Editor-agnostic (E-001) |

+ | `packages/editor` | The editing experience: tools, `useEditor`, panels, floor plan, paint, shortcuts, overlays |

+ | `packages/nodes` | The built-in plugin `pascal:core`: one folder per node kind with its definition (E-002) |

+ | `packages/mcp` | MCP server, scene storage adapters, agent tools |

+ | `packages/cli` | Packed CLI + portable runtime |

+ | `packages/ifc-converter` | IFC import/export |

+ | `apps/editor` | Standalone app composing viewer + editor |

+ | `skills/` | Public agent skills shipped with the plugin (product payload, validated in CI) |

## Where to look

- - **Architecture rules** — `wiki/architecture/` (read on demand; index in `wiki/architecture/README.md`).

- - **Skills (ready workflows)** — `.agents/skills/<name>/SKILL.md`. Same content is reachable as `.claude/skills/`, `.cursor/skills/`, `.codex/skills/` (symlinks to `.agents/skills/`).

- - **Repo orientation for humans** — `README.md`, `SETUP.md`, `CONTRIBUTING.md`.

- `CLAUDE.md`, `GEMINI.md`, and `.github/copilot-instructions.md` are symlinks to this file. Codex reads this file directly.

- ## Layer Boundaries (read once, internalise)

- - **`packages/core`** owns domain data and pure logic. It must not import Three.js, `packages/viewer`, `apps/editor`, rendering/UI concepts, tools, modes, phases, or view-specific concepts such as floorplan or paint preview.

- - **`packages/viewer`** owns the standalone 3D canvas, renderers, viewer systems, and genuine presentation state. It must not know about `useEditor`, editor tools, phases, modes, paint mode, floorplan state, or editor-only presentation vocabulary.

- - **`apps/editor`** owns the editing experience: tools, `useEditor`, panels, floorplan helpers, paint mode, keyboard shortcuts, command palette, action menus, cursor badges, and editor-only overlays. Editor features are injected into `<Viewer>` via props and children.

- Details, examples, and rationale live in `wiki/architecture/layers.md`, `wiki/architecture/viewer-isolation.md`, `wiki/architecture/systems.md`, `wiki/architecture/renderers.md`, `wiki/architecture/tools.md`.

- ## When making architecture-sensitive changes

- Read the relevant page in `wiki/architecture/` **before** writing code. The page list lives in `wiki/architecture/README.md`. As a minimum:

+ - Architecture detail, on demand: `wiki/architecture/` (index in its `README.md`). Read the page for the boundary you touch: a new kind → `node-definitions.md`, `node-schemas.md`; a tool or interaction → `tools.md`, `interaction-scope.md`; anything in `viewer` → `viewer-isolation.md`, `layers.md`; selection → `selection-managers.md`.

+ - Reviewing a PR: `.agents/skills/review-architecture/SKILL.md`.

+ - Opening a PR: `.agents/skills/open-pr/SKILL.md`.

+ - Humans: `README.md`, `SETUP.md`, `CONTRIBUTING.md`.

- - Adding a node type → `node-schemas.md`, `renderers.md`, `systems.md`

- - Adding a tool → `tools.md`, `spatial-queries.md`, `events.md`

- - Adding / changing a placement or move interaction → `tools.md` ("2D ↔ 3D behavioral parity": applicable behaviors must exist in both views; port the change to the sibling 2D/3D file in the same PR)

- - Adding / changing an MCP tool or resource, or the `pascal-3d` skill → `agent-surfaces.md` (capability and knowledge parity with the hosted AI chat: port the change to the other surface in the same PR, or record why it doesn't apply)

- - Adding a system → `systems.md`, `scene-registry.md`

- - Anything in `packages/viewer` → `viewer-isolation.md`, `layers.md`

- - Anything touching selection → `selection-managers.md`, `scene-registry.md`, `events.md`

+ `CLAUDE.md`, `GEMINI.md` and `.github/copilot-instructions.md` symlink here.

- ## When reviewing a PR

+ ## Commands

- Invoke the `review-architecture` skill (`.agents/skills/review-architecture/SKILL.md`). It loads the required architecture pages, fetches the diff, classifies each new file by layer, and reports findings grouped by severity.

+ `bun run ci` is exactly what CI runs (check, skills:validate, check-types, test, build). The git hooks (installed by `bun install`) are fast feedback: Biome on staged files, typecheck of the packages you changed.

- ## Operating rules

+ ## Working agreements

- - Read the full file before editing. Plan all changes, then make one complete edit.

- - When the user corrects you, stop and re-read their message.

- - After two consecutive tool failures, stop and change approach.

- - Don't introduce backwards-compatibility shims, dead code, or speculative abstractions.

- - Don't write new comments unless they explain a non-obvious *why*.

+ - Rewrite when the shape is wrong; say what you are replacing and why; keep the PR to its stated goal.

+ - No compatibility shims, dead code or speculative abstractions; comments only for a non-obvious *why*.

+ - Old scenes must still load after your change (E-003).

+ - Never cite private-repo plan paths from this repo (E-015).