Before calling any skill in this module: if you are about to call a skill with parameters guessed from its name or description, STOP — read this file (or fetch its schema via GET /skills/recommend?includeSchema=true) first. If you already have the parameter definitions from recommend/schema, you may proceed straight to dryRun.

Triggers

• Framing the Scene View
• Creating or adjusting cameras
• Tweaking FOV or clipping planes
• 取景 Scene View、创建或调整相机、修改 FOV 或裁剪面

Camera Skills

Control the Scene View camera and Game Cameras (creation, transform, properties, screenshot, culling, orthographic toggle).

Operating Mode

• Approval (default): mutating skills (camera_set_transform, camera_create, camera_set_properties, camera_set_culling_mask, camera_screenshot, camera_sceneview_screenshot, camera_set_orthographic, camera_align_view_to_object, camera_look_at) need user grant; grant triggers a single server-side execution that returns the result.
• Auto / Bypass: those skills execute directly.
• Query skills (camera_get_info, camera_get_properties, camera_list) are SkillMode.SemiAuto — they run in all three modes without grant.
• This module contains no Delete / PlayMode / Reload / high-risk skills (no NeverInSemi).

Guardrails

DO NOT (common hallucinations):

• camera_move / camera_rotate do not exist → use camera_set_transform (Scene View) or gameobject_set_transform (Game Camera)
• camera_set_fov does not exist → use camera_set_properties with fieldOfView parameter
• camera_* skills control two different cameras: camera_set_transform/camera_look_at/camera_align_view_to_object control the Scene View camera; camera_create/camera_set_properties/camera_screenshot control Game Cameras
• camera_delete does not exist → use gameobject_delete on the camera GameObject

Routing:

• For Cinemachine virtual cameras → use cinemachine module
• For Game Camera component properties → camera_set_properties / camera_get_properties (this module)
• For screenshots → three options: scene_screenshot (scene module) = the Game View final composite (all cameras + UI; Play mode = live runtime frame); camera_screenshot (this module) = a single Game Camera off-screen render; camera_sceneview_screenshot (this module) = the editor Scene View (developer viewport, incl. grid/gizmos)

Skills

camera_align_view_to_object

Align Scene View camera to look at an object. Parameters:

• name (string, optional): Target GameObject name.
• instanceId (int, optional): Target GameObject instance ID.
• path (string, optional): Target GameObject hierarchy path.

camera_get_info

Get the editor Scene View viewport camera's position and rotation — the developer's viewport, not a Camera component in the scene. It takes no target parameters for exactly that reason. For a Game Camera's state use camera_get_properties (or camera_list to find one); if you just created a camera with camera_create and want to confirm its transform, camera_get_info will not show it. Parameters: None.

camera_set_transform

Set Scene View camera position/rotation manually. Parameters:

• posX, posY, posZ (float): Position.
• rotX, rotY, rotZ (float): Rotation (Euler).
• size (float, optional): Orthographic size or pivot distance (default 5).
• instant (bool, optional): Move instantly (default true).

camera_look_at

Focus Scene View camera on a world-space point. Parameters:

• x, y, z (float): Target point.
• Does not support targetName or GameObject lookup. For object focus, use camera_align_view_to_object.

camera_create

Create a new Game Camera.

Returns: { success, name, instanceId }

camera_get_properties

Get Game Camera properties (supports name/instanceId/path).

Returns: { success, name, fieldOfView, nearClipPlane, farClipPlane, orthographic, orthographicSize, depth, cullingMask, clearFlags, backgroundColor, rect }

camera_set_properties

Set Game Camera properties (FOV, clip planes, clear flags, background color, depth).

Any bg* channel you omit keeps the camera's current value, so you can set alpha alone.

Returns: { success, name, applied, fieldOfView, nearClipPlane, farClipPlane, orthographic, orthographicSize, depth, cullingMask, clearFlags, backgroundColor, rect } — the response echoes the camera's state after the write, and applied lists which parameters were actually changed, named exactly as you passed them (the background channels appear individually as bgR, bgG, bgB, bgA, not lumped under backgroundColor), so a parameter absent from applied was not supplied. That echo is your verification: a second camera_get_properties call is redundant. An unparseable clearFlags rejects the whole call with SEMANTIC_INVALID + validValues before anything is applied, so the numeric parameters in the same call are never half-written.

camera_set_culling_mask

Set Game Camera culling mask by layer names (comma-separated).

Returns: { success, cullingMask }

camera_screenshot

Capture a screenshot from a Game Camera to file.

Returns: { success, path, width, height }, plus { imageBase64, imageWidth, imageHeight, imageBytes } when returnImage=true. If the base64 payload would exceed 8MB, the skill returns an error asking for a smaller maxDimension — the file at path is still saved.

camera_sceneview_screenshot

Capture the editor Scene View (the developer's editing viewport — can overlook the whole scene incl. off-camera objects). Distinct from scene_screenshot (Game View / player camera) and camera_screenshot (one Game Camera). By default captures the full Scene View incl. grid/gizmos/selection (on-screen read); auto-falls back to a clean offscreen render if the editor build lacks the internal API. The Scene View window must be open and visible for the overlay capture.

Returns: { success, path, width, height, mode, note } — mode is "screen_with_overlays" or "offscreen_clean" — plus { imageBase64, imageWidth, imageHeight, imageBytes } when returnImage=true. If the base64 payload would exceed 8MB, the skill returns an error asking for a smaller maxDimension — the file at path is still saved.

returnImage usage tip: a local agent that can read files (e.g. Claude Code against a local Unity Editor) should generally omit returnImage and just read the PNG at path — it's cheaper on tokens. Use returnImage=true for remote/MCP clients that have no filesystem access to the Unity project.

camera_set_orthographic

Switch Game Camera between orthographic and perspective mode.

Returns: { success, orthographic, orthographicSize }

camera_list

List all cameras in the scene.

Returns: { count, cameras: [{ name, instanceId, path, depth, orthographic, enabled }] }

Exact Signatures

Exact names, parameters, defaults, and returns are defined by GET /skills/schema or unity_skills.get_skill_schema(), not by this file.