evomeme 0.1.9 → 0.1.10-beta.13

package/dist/skills/en-US/game-asset-batch-generator/SKILL.md

@@ -0,0 +1,117 @@
+---
+name: game-asset-batch-generator
+description: Use evomeme_generate_asset_batch to generate one consistent static game asset sheet in top-down, isometric, side or custom view, optionally remove its background color and always split it into independent PNGs. Use for scene-building packs of buildings, vegetation, props, walls, floor tiles or platforms. Explains processing, actual outputs, view/color-key/splitting terms and game integration. Not for splitting an existing sheet only, a single background, animation or guaranteed seamless tiles.
+---
+
+# Batch game asset generation
+
+Use this skill for groups of static 2D assets sharing a style and viewpoint for scene assembly. Establish the theme, view, style references and background mode, call `evomeme_generate_asset_batch`, then inspect the separated images before delivery.
+
+## What the tool actually does
+
+MCP and the canvas batch feature share the `generate_asset_batch` task and `executeAssetBatchGeneration` function. The shared task layer sequences existing image functions. No MCP calls another MCP, and no canvas feature entry point chains another canvas entry point. Shared task-layer orchestration is sufficient; do not introduce another workflow or change built-in image prompts merely to port the tool.
+
+One call performs these steps:
+
+1. **One AI image generation**: Generate one sheet of separate objects from the theme, view, optional references and existing asset-category requirements. It does not send a separate generation request per object or produce one image per reference.
+2. **Download the sheet**: Save and retain the original generation under workspace `assets/`.
+3. **Process the chosen background mode**: By default, generate a solid background, remove it using color key at fixed strength 80, and save a transparent PNG. `transparent_background=true` requires Max, requests transparency directly and skips color key.
+4. **Always split automatically**: Crop detected independent foreground regions into PNGs. Subjects are not resized or padded to equal dimensions. There is no input to disable splitting.
+
+Successfully splitting N subjects produces **N independent PNGs (N is at least 2)**. Retain one original sheet and, when color-keyed, one additional keyed sheet. A normal successful call therefore adds N+1 or N+2 images, excluding existing reference assets. Read the original sheet format from `mime_type`; keyed and split outputs are PNG.
+
+Counts depend on generated content and foreground detection; requesting a number of objects does not guarantee that many files. Outputs are image resources, not complete levels, object coordinates, colliders, engine prefabs, atlas-coordinate configuration, animation sequences or a guarantee of seamless tiling.
+
+## When to use it
+
+| Need | Choice |
+| --- | --- |
+| A themed scene-building pack of buildings, vegetation, props and walls | Use this tool with a consistent view and art reference. |
+| Platforms, stairs, slopes and decorations for a side-scrolling level | Use `side-view`, then verify that objects retain a strict side view. |
+| One object or one complete background | Use [general image generation](../game-asset-image-generator/SKILL.md) to avoid mandatory splitting. |
+| Seamless repeating tiles or transition atlases between two terrains | Use the [tile-map skill](../game-tile-generator/SKILL.md). Floor tiles from this batch tool are not guaranteed seamless. |
+| Looping scroll backgrounds, midgrounds and foregrounds | Use the [scroll-map skill](../game-scroll-generator/SKILL.md). |
+| Button/panel skins and their runtime use | Use the [UI component skill](../game-ui-generator/SKILL.md). |
+| Extract objects from an existing sheet only | Use existing image-processing capabilities; this tool generates new art and is unsuitable for splitting only. |
+
+## Terms and view selection
+
+- **Asset sheet**: One image containing multiple static objects separated by space for cropping. It is not an assembled game scene or a time-ordered animation sheet.
+- **Foreground subjects and automatic splitting**: Foreground means the object regions to retain. Pixel, transparency and connected-region detection separate subjects without semantic recognition such as “tree” or “chest.” A connected region is a group of adjacent pixels. Touching objects can merge and detached decorations can split incorrectly. This is neither fixed-grid cutting nor layer decomposition.
+- **Color key**: Make a selected background color and similar colors transparent. The existing workflow uses Oklab perceptual color distance, which compares colors more closely to human perception, at fixed strength 80 over the whole image with soft edges. Similar colors inside objects may also disappear; it does not infer object contours like semantic background removal.
+- **Native transparency / alpha channel**: The model directly generates transparent regions; alpha records pixel opacity. Transparent regions should reveal the scene behind them, rather than painted white or a checkerboard. Consider Max native transparency for glass, glows and soft shadows, and inspect the result.
+- **Despill**: Remove residual key-color tint from object edges. It is disabled in this workflow, so inspect magenta fringes, glows and translucent details. This tool cannot change that setting.
+- **Equal-size padding**: Add transparent pixels around crops so all images have identical dimensions. This workflow fixes `pad_to_same_size=false`; output sizes usually differ and subjects are not resized to a common size.
+- **Anchor / pivot**: The point used to position an image in the game, such as a tree root or a building's ground-contact center. Splitting internally uses `anchor=center`, but with padding disabled this does not configure engine ground points, pivots or colliders. Set those after inspecting actual images.
+
+`asset_type` selects the **viewpoint**, not object category or file format:
+
+| Value | Appearance, use and existing requested content |
+| --- | --- |
+| `top-down` (default) | Look down from above, suitable for roofs, ground objects, vegetation and props in overhead maps. Built-in categories include floor tiles, walls, fences, vegetation, buildings, props and atmospheric decorations. |
+| `isometric` | An orthographic oblique, isometric-style view usually showing tops and sides, with consistent apparent scale at different distances. Suitable for oblique scene assembly. The existing preset says “orthographic 45-degree isometric view”; this is an image-generation constraint, not a verified 3D camera angle or exact grid. Categories match top-down. |
+| `side-view` | View objects from the side for side-scrolling scenes. In addition to buildings and vegetation, request floating platforms, side walls, ascending/descending stairs and slopes. It is not a character-direction sheet or movement animation. |
+| `custom` | Describe the viewing direction in `custom_asset_type`, e.g. “front orthographic view of building facades.” Common categories are still appended; this is not a category-list replacement or exact-count interface. |
+
+References guide style, colors, materials and design, without overriding the selected viewpoint or separate-layout requirements. Every mode retains existing built-in categories; the user `prompt` adds theme and content requirements without guaranteeing an exclusive subset.
+
+## Parameters and calls
+
+| Parameter | Actual meaning |
+| --- | --- |
+| `prompt` | Required, nonempty. Describe theme, style, colors, materials, desired content and relative proportions. |
+| `asset_type` | Optional; defaults to `top-down`. Other modes are explained above. |
+| `custom_asset_type` | Required nonempty view description for `custom`; unused in other modes. |
+| `model` | Default `asset-pro`; also `asset-max` or `asset-lite`. Only Max supports native transparency in this workflow. |
+| `size` | Whole-sheet preset, default `1664x2496`, not each asset's dimensions. Must fit model limits; Lite does not support `1024x1024`. |
+| `reference_ids` | Workspace reference-image IDs, default `[]`. Up to 14 in this workflow, with Pro limited to 10. |
+| `background_color` | Default `#ff00ff`; accepts `#RGB` or `#RRGGBB`. Solid background/key color; avoid subject colors. Not used as the background with native transparency. |
+| `transparent_background` | Default `false`: generate solid background then color-key. True requests Max native transparency; automatic splitting still runs. |
+| `workspace_dir` | Absolute workspace root; specify explicitly when possible. Omission uses the MCP working directory. |
+
+There are no `count`, `components`, `rows`, `columns`, `auto_split`, `pad_to_same_size` or color-key-strength inputs. Do not invent or pass them. Fixed splitting behavior is described in the results.
+
+Place reference files under the chosen workspace's `assets/`, verify readability and avoid overwriting existing assets. Download accessible external URLs into the workspace before passing IDs such as `assets/style.png`. Parameters do not accept URLs, Base64 or original attachment absolute paths. Reuse an existing `asset_id` when appropriate.
+
+```json
+{
+  "workspace_dir": "D:/my-game",
+  "prompt": "A warm hand-painted medieval village asset pack, wood and pale stone bricks, ochre and green palette, consistent upper-left lighting. Buildings, trees, fences, crates and roadside decorations have coordinated proportions. Show every object completely with wide gaps and clear silhouettes; no continuous ground or complete scene, and avoid magenta on subjects.",
+  "asset_type": "isometric",
+  "model": "asset-pro",
+  "size": "1664x2496",
+  "reference_ids": [],
+  "background_color": "#ff00ff",
+  "transparent_background": false
+}
+```
+
+Example with a custom viewpoint and native transparency:
+
+```json
+{
+  "workspace_dir": "D:/my-game",
+  "prompt": "Fairytale town scene-building assets, consistent soft palette and flat illustration style, complete separate building facades, flowerpots, streetlights and decorations, retain fine silhouettes, no text or complete scene.",
+  "asset_type": "custom",
+  "custom_asset_type": "Front orthographic view, no overhead view or side perspective",
+  "model": "asset-max",
+  "transparent_background": true
+}
+```
+
+## Results and use
+
+MCP returns text metadata with arrays/objects serialized as JSON text, without image bytes or Base64:
+
+- `images` contains final split PNGs with `id` / `asset_id`, `name`, `kind`, `url`, `operation_id`, `local_path` and `mime_type`. Top-level `asset_id` / `local_path` describe the first image, not the entire pack. A `/files/...` URL belongs to the local canvas service, not a public asset host.
+- `generated` describes the original sheet with its remote `job_id` and `tos_url`. When color key runs, `color_key` describes the keyed sheet; native transparency may omit it. For a sheet preview, prefer `color_key`, otherwise `generated`.
+- `asset_type`, applicable `custom_asset_type`, background settings, `auto_split=true` and `pad_to_same_size=false` describe processing. `color_key_strength=80` applies only to the color-key path.
+- `operation_id` identifies a local step; the sheet's `job_id` identifies the remote AI generation. Existing shared results do not provide `task_id`, `outputs`, crop boxes or actual dimensions; crop boxes and output sizes are saved in the splitting-step record.
+
+Filenames use a batch identifier and numeric suffixes such as `_asset_01.png`, not semantic names. Inspect the sheets and individual PNGs, build a file-to-purpose list from actual content, check touching objects, incorrect splits, missing items, background residue and style consistency, then rename as needed.
+
+Use individual PNGs as static sprites/textures in the game. A sprite is an object displaying a 2D image. Set pivots from ground-contact points, scale and place assets using a consistent world scale, and configure draw order and colliders separately. Image bounds do not define collision shapes. Verify scale consistency, grid alignment and seamlessness separately. If an atlas is needed, pack assets using the project's existing tools and retain trimming/pivot data.
+
+Final keyed and split images remain local and are not uploaded automatically. For public URLs, upload the final `local_path` through `evomeme_upload_game_asset`; do not substitute the original sheet URL for a separated object.
+
+Splitting needs at least two separable subjects. On failure, read the error and locate the generated sheet, optional keyed sheet and any written files through task records or the canvas; failure does not guarantee a complete result list. Do not immediately pay to regenerate because splitting failed. Inspect existing art and choose cropping/processing first; adjust spacing, background or prompts if regeneration is actually needed.

package/dist/skills/en-US/game-asset-batch-generator/agents/openai.yaml

@@ -0,0 +1,7 @@
+interface:
+  display_name: "Batch Game Asset Generation"
+  short_description: "Generate game asset packs with view, keying and split guidance"
+  default_prompt: "Use $game-asset-batch-generator to create a game asset pack in a chosen view and explain its processing, output files and game integration."
+
+policy:
+  allow_implicit_invocation: true

package/dist/skills/en-US/game-asset-image-generator/SKILL.md

@@ -0,0 +1,86 @@
+---
+name: game-asset-image-generator
+description: Generate game concept art, backgrounds and image assets through MCP, or apply smart cutout and background-color removal to existing images. Explains text-to-image, image-to-image, transparent PNG outputs and reference preparation. Use dedicated skills for character views and videos.
+---
+
+# Generate and Process Game Images
+
+For character four views or eight directions, use [Character View Generator](../game-character-view-generator/SKILL.md). For videos, use [Game Video Generator](../game-video-generator/SKILL.md). Always use [Tilemap Resource Generation, Editing and Optimization](../game-tile-generator/SKILL.md) when generating tilemaps or seamless tiles; also use it for transition tilesets and existing-tile seam repair.
+
+## Workspace and Outputs
+
+Always use [Scrolling Map Resource Generation and Editing](../game-scroll-generator/SKILL.md) for scrolling maps, horizontally looping backgrounds, parallax layers and left/right seam repair of existing scrolling images.
+
+Explicitly pass the absolute current workspace root as `workspace_dir`. Outputs are saved under its `assets/` directory. Record the returned `asset_id`, `local_path` and remote URL.
+
+Platform games read TOS URLs through project configuration, usually `public/config.json`. Color removal produces a local PNG. Before using it in a platform game, upload its `local_path` with `evomeme_upload_game_asset` and record the returned URL.
+
+## Prepare References Before Invoking Generation
+
+When using an uploaded attachment, conversation image, concept image or other local reference, you MUST place it under `assets/` in the selected `workspace_dir` before calling a generation MCP tool:
+
+1. Locate the actual attachment file provided by the IDE and confirm it is readable. If only an accessible image URL is available, download the image first. If the actual file cannot be obtained, explain the missing source and stop that reference-based call; do not invent a path or asset ID.
+2. Reuse images already under this workspace's `assets/`. Copy images from any other location, including other workspace directories, into `assets/` or a subdirectory such as `assets/references/`. Use a new filename for different images with conflicting names to avoid overwriting existing assets.
+3. Confirm the copied image exists and is readable. Its asset ID is the path relative to the workspace root, using `/` separators. For example, `D:/my-game/assets/references/hero.png` maps to `assets/references/hero.png`.
+4. Pass these IDs through `reference_ids` for ordinary image-to-image generation, `character_reference_id` for four views or eight directions, and `four_view_id` for supplied four views. Explicitly pass the same `workspace_dir`. These fields do not accept original attachment absolute paths, URLs or image Base64.
+
+Reuse `asset_id` returned by existing generation outputs without copying again. This makes references workspace assets, allowing tasks to record provenance and the canvas to draw dependency connections. Text-to-image generation without references needs no reference-file preparation.
+
+## Three Peer Image Tools
+
+| Tool | Action and generated output | When to use |
+| --- | --- | --- |
+| `evomeme_generate_game_image` (Image Generation) | Generate one new image from a prompt and optional references; save it under `assets/` and return its path and cloud URL. Read the actual format from `mime_type`. | Create concept art, characters, props or backgrounds. No references means text-to-image; supplied references mean image-to-image. |
+| `evomeme_cutout_game_image` (Smart Cutout) | Identify the subject, generate a mask, then compose, save and upload one transparent PNG. The mask specifies how much of each pixel to retain or make transparent. | Keep a character or prop while removing its background, especially when the background has multiple colors. |
+| `evomeme_color_key` (Color Removal) | Locally make pixels matching a specified color transparent and save one new PNG under `assets/`; no automatic upload. | Smart cutout is unavailable and the background color is known, such as solid white or magenta. |
+
+All three tools share their client functions with the corresponding canvas features. A transparent PNG records transparency in its alpha channel so it can overlay a game scene; transparency is not a white-painted background.
+
+Independent asset tasks can run in parallel. Cutout or color removal depends on the generated image, so execute those steps after generation. Image generation does not automatically cut out the result.
+
+## Image Generation
+
+| Parameter | Required | Description |
+| --- | --- | --- |
+| `prompt` | Yes | Generation prompt. |
+| `model` | No | `asset-max`, `asset-pro`, or `asset-lite`; default `asset-pro`. |
+| `background` | No | `opaque`, `transparent`, or `auto`; default `opaque`. |
+| `size` | No | Preset supported by the selected model; default `2048x2048`. |
+| `reference_ids` | No | Array of workspace reference image asset IDs, such as `["assets/hero.png"]`. Omit or pass an empty array for text-to-image. |
+| `workspace_dir` | No | Absolute current workspace root; pass it explicitly. |
+
+Presets include `1024x1024`, `2048x2048`, `1728x2304`, `2304x1728`, `2560x1440`, `1440x2560`, `1664x2496`, `2496x1664`, and `3024x1296`. `asset-lite` does not support `1024x1024`.
+
+For transparent text-to-image, use `model="asset-max", background="transparent"`. Auto background also requires `asset-max`. `asset-lite` does not offer background switching. For `asset-pro` transparent generation, `reference_ids` must contain exactly one PNG with transparent pixels. Reference IDs are relative to `workspace_dir` and accept `asset_id` returned by image tools. Maximum references: `asset-pro` 10, `asset-lite` 14, `asset-max` 16. Parameters and reference handling match canvas image generation, except MCP accepts only preset sizes and no custom sizes or legacy image-type parameters.
+
+Read the [Static Game Element Prompt Guide](references/static-game-assets.md) for individual static elements or the [Game Background Prompt Guide](references/background.md) for full backgrounds. If concept art exists, inspect it and describe its shapes, palette and style in the prompt. Supply workspace images through `reference_ids` when direct references are needed.
+
+## Smart Cutout
+
+Provide exactly one of `id` or `image_url`, plus `workspace_dir`:
+
+- `id`: Workspace image asset ID such as `assets/hero.png`; use the generator's returned `asset_id`.
+- `image_url`: Accessible HTTP/HTTPS source image URL.
+
+The tool waits for the mask, composes a transparent PNG locally and uploads it. Errors are passed through to MCP; it does not switch to color removal automatically.
+
+## Color Removal
+
+| Parameter | Required | Description |
+| --- | --- | --- |
+| `id` | Yes | Workspace source image asset ID. |
+| `color` | No | Color to remove, `#RRGGBB`; default `#ffffff`. |
+| `strength` | No | 0–100; default 80. |
+| `contiguous_only` | No | Remove only regions connected to an image boundary; default `false`. |
+| `despill` | No | Remove background-color spill from retained edges; default `false`. |
+| `workspace_dir` | No | Absolute current workspace root. |
+
+This runs locally without authentication or the smart-cutout service. Select the removal color from the source background. It removes matching colors rather than recognizing a semantic subject.
+
+`strength` controls color-match tolerance: higher values remove more nearby colors. With `contiguous_only=true`, only matching regions reachable from image boundaries are removed, reducing holes in same-colored areas inside the subject. `despill` reduces background color mixed into retained translucent edges; it does not recognize or redraw the subject.
+
+## Results and Errors
+
+All three tools return only text metadata on success, without image Base64 or MCP `image` content blocks. Text includes `asset_id`, `local_path`, `mime_type` and processing details. Generation and smart cutout also return `job_id`, `asset_url`/`tos_url`, `status` and `attempts`. Color removal returns a local `operation_id` and its four processing parameters. Use `asset_id` for further processing and URLs to reference cloud images. To inspect an image, use an available IDE image viewer to read `local_path` or open the image URL as needed; do not put image Base64 into the conversation or project configuration.
+
+Failures carry `isError: true` and the specific error message. Keep the original image, record the error, and select color-removal settings from its background when smart cutout is unavailable.

package/dist/skills/en-US/game-asset-image-generator/agents/openai.yaml

@@ -0,0 +1,7 @@
+interface:
+  display_name: "Game Asset Image Generator"
+  short_description: "Generate and process game image assets, save the results to the workspace, and connect them to the project configuration"
+  default_prompt: "Please use $game-asset-image-generator to generate or process the image assets needed for my game, and correctly connect the files and asset links to the project."
+
+policy:
+  allow_implicit_invocation: true

package/dist/skills/en-US/game-asset-image-generator/references/background.md

@@ -0,0 +1,34 @@
+# Game Background Prompt Guide
+
+This guide covers ordinary complete single-layer backgrounds. Always use [Scrolling Map Resource Generation and Editing](../../game-scroll-generator/SKILL.md) for scrolling maps, horizontal looping backgrounds, parallax layers or scroll seam repair; that workflow handles optional layers, seam repair and color key processing.
+
+A game background is a complete single-layer scene image that fills the game screen, such as a forest, city street, outer-space area, underground cave, or indoor space. Each background is generated only once and used as a separate `background` image.
+
+## Use Concept Art As Reference
+
+If concept art exists for the game or current scene, inspect it and describe the relevant features in the prompt. Before directly referencing an image, you MUST copy it into this workspace's `assets/` or reuse it if already there, verify it is readable, and pass its workspace-relative ID through `reference_ids` with the same explicit `workspace_dir`:
+
+- Prefer the setting concept image for the current scene; if none exists, then use the overall game concept image.
+- Clearly state in the prompt which characteristics need to be inherited, such as worldview, scene structure, color palette, materials, lighting, and art style.
+- Do not copy layout, text, annotations, character presentation, or other irrelevant content from the concept image.
+
+## Prompt
+
+When writing, follow these rules:
+
+- Clearly state the background's purpose in the game, scene location, era, or theme.
+- Clearly state landscape or portrait composition, camera perspective, horizon position, and main spatial direction.
+- Clearly state time, weather, lighting, main color palette, and main materials, and keep them consistent with existing game art specifications.
+- Specify areas that need to remain clear and low-interference according to the gameplay, such as the central movement area, road, platform area, or character activity area.
+- Place high detail and visual focus where they will not block gameplay information, and avoid uniform detail density across the whole image.
+- Keep spatial layering, perspective, light direction, and scale consistent so the background can directly fill the game screen.
+- Require `full-frame composition, edge-to-edge scene, no border`; do not generate a white background, transparent background, picture frame, or display mockup.
+- End with `no characters, no enemies, no pickups, no UI, no text, no watermark, no logo`.
+
+## Prohibited Wording
+
+- Drawing characters, enemies, props, pickups, attack effects, or interactive objects into the background.
+- Drawing HUD, buttons, title, score, tutorial text, Logo, or watermark into the background.
+- Describing only atmosphere without specifying the concrete scene, perspective, composition direction, and gameplay negative space.
+- Requesting standalone elements, transparent background, pure white background, asset collections, setting sheets, or multi-view images.
+- Using an art style, color palette, materials, and lighting that conflict with the game concept image or existing art specifications.

package/dist/skills/en-US/game-asset-image-generator/references/static-game-assets.md

@@ -0,0 +1,38 @@
+# Static Game Element Prompt Guide
+
+A static game element is a single game element that can be directly represented by one standalone image and does not need animation frames or dynamic effects, such as a small stone, standalone cloud, shrub, road sign, or an enemy that only uses a fixed pose. Each static element is generated only once and used as a separate `asset` image.
+
+This guide does not apply to:
+
+- Complete scene backgrounds, parallax background layers, or tilesets.
+- Characters or enemies that need animation frames such as walking, attacking, or taking hits.
+- Dynamic effects such as fire, explosions, or trails.
+- Game concept art, character setting art, UI, videos, or 3D models.
+
+## Use Concept Art As Reference
+
+If concept art exists for the game or current element, inspect it and describe the relevant features in the prompt. Before directly referencing an image, you MUST copy it into this workspace's `assets/` or reuse it if already there, verify it is readable, and pass its workspace-relative ID through `reference_ids` with the same explicit `workspace_dir`:
+
+- Prefer the setting concept image for the current element; if none exists, then use the overall game concept image.
+- Clearly state in the prompt which characteristics need to be inherited, such as element shape, silhouette, color palette, material, and art style.
+- Do not copy layout, text, annotations, multi-view presentation, or other irrelevant elements from the concept image.
+
+## Prompt
+
+When writing, follow these rules:
+
+- One image describes only one static element; collections, multiple variants, or multi-view images must not be requested.
+- Clearly state the element's role in the game, such as obstacle, fixed enemy, or environmental decoration.
+- For directional elements, the single facing direction must be specified; for non-directional elements, specify front view, side view, or top view.
+- For a fixed enemy, describe only one pose that can be used directly.
+- Keep only one to three key identifying features to ensure the silhouette remains clear after scaling down.
+- Require the subject to be centered, fully visible, surrounded by whitespace, and not cropped at the edges.
+- Use `pure white background, no ground, no cast shadow, clean edges` to support subsequent smart cutout or white-color removal.
+- End with `no text, no watermark, no logo, no UI, no extra objects`.
+
+## Prohibited Wording
+
+- `background scene`, `parallax layer`, `tileset`.
+- `character sheet`, `multiple views`, `sprite sheet`, `animation frames`, `turnaround sheet`.
+- Multiple elements, multiple poses, multiple directions, or multiple skins appearing in one image at the same time.
+- Scene narrative, ground, sky, environmental lighting effects, or other content that would interfere with independent cutout.

package/dist/skills/en-US/game-asset-mesh-generator/SKILL.md

@@ -0,0 +1,35 @@
+---
+name: game-asset-mesh-generator
+description: Generate and postprocess Three.js GLB assets through evomemeMCP (evomeme_text_to_mesh, evomeme_image_to_mesh, and related tools). Use for text/image-to-3D, texturing, segmentation, conversion, rigging, animation, or TOS-hosted model URLs.
+---
+
+# game-asset-mesh-generator
+
+Generate and process 3D assets through the local **evomemeMCP** tools. Local tasks manage execution and downloads; the remote plugin server handles model services and credentials. Use the MCP tools for generation and remote processing.
+
+## Choose the relevant reference
+
+Read [tool-reference.md](references/tool-reference.md) for tool selection, common task options, returned files/IDs and recovery. Then read only the reference needed for the operation:
+
+- Text, single-image or multiview generation: [generation-contracts.md](references/generation-contracts.md).
+- Texture, stylize, segment, complete, decimate, convert, rig or retarget: the selected section of [postprocess-contracts.md](references/postprocess-contracts.md).
+- Animation selection: [animation-presets.md](references/animation-presets.md).
+- Concept-image preparation: [image-generator-workflows.md](references/image-generator-workflows.md).
+
+## Workflow
+
+1. Use the request's asset role, style, scale, performance budget and animation needs. Clarify only details that affect the result and cannot be inferred.
+2. Choose the generation or processing tool from the tool reference. Upload local source files with `evomeme_upload_game_asset` when a remote input is needed; use its public URL. A disk path or `file://` URL is not a model-service input.
+3. Apply the selected operation's parameter constraints. Generation family selection and postprocess versions are documented separately; do not carry generation options into other operations.
+4. Retain the returned files, IDs and full result. Follow the readiness and recovery rules in the tool reference before chaining operations or importing assets.
+5. For game integration, use the downloaded `local_path` and follow [game-asset-align](../game-asset-align/SKILL.md): CLI import → review → Stage user confirmation → sync. See its [CLI contract](../game-asset-align/references/tool-contract.md) and [Three.js integration](references/threejs-integration.md). Convert incompatible formats before importing.
+
+## Character workflow
+
+Generate a fused full-body character in a clear T-pose/A-pose, or the stance required by the chosen creature preset. Select the body plan in `rig_type`, rig the model, then retarget using the returned **rig task ID**. Read the animation presets before choosing clips. Inspect the skeleton and animation before deciding whether another paid rigging attempt is needed.
+
+MCP callers use atomic generation → rig → retarget tools. Character composition is internal to `tasks`; there is no character-pipeline or rig-check MCP tool.
+
+## Completion evidence
+
+Report the source, final local files and URLs, task IDs, actual format, generation/quality settings and any remaining issue. For imported assets, include available triangle counts, materials/textures, scale, pivot, bounds and animation clips. Claim completion only for stages whose results have been verified. If a tool fails, preserve its literal error and existing task identifiers.

package/dist/skills/en-US/game-asset-mesh-generator/agents/openai.yaml

@@ -0,0 +1,7 @@
+interface:
+  display_name: "Game Asset Mesh Generator"
+  short_description: "Generate and postprocess 3D mesh assets (text/image-to-3D, texture, rig, animation)"
+  default_prompt: "Use $game-asset-mesh-generator to generate or postprocess the required GLB assets through the remote MCP."
+
+policy:
+  allow_implicit_invocation: false

package/dist/skills/en-US/game-asset-mesh-generator/references/animation-presets.md

@@ -0,0 +1,313 @@
+# Animation presets
+
+Pass to `evomeme_retarget_mesh` via `animation` (one preset) or `animations` (preset array; legacy comma-separated string also accepted). Explicit strings/arrays pass through unchanged; only legacy comma strings and the omitted walk default are normalized using the rig-version hint. Match `rig_type` to the preset family.
+
+| `rig_type` | Rig model | Catalog |
+| --- | --- | --- |
+| `biped` | `v2.5-20260210` | `preset:idle`, `preset:walk`, `preset:run`, `preset:dive`, `preset:climb`, `preset:jump`, `preset:slash`, `preset:shoot`, `preset:hurt`, `preset:fall`, `preset:turn` |
+| `biped` | `v1.0-20240301` | Large `preset:biped:*` library below |
+| `quadruped` / `hexapod` / `octopod` / `serpentine` / `aquatic` / `avian` | `v2.5-20260210` | One locomotion each (except avian) |
+
+The official retarget contract does not specify a maximum batch size or guaranteed clip names/order. Preserve the complete result and inspect the downloaded animation clips; do not automatically split into extra paid calls based on an assumed five-preset limit.
+
+Generate creatures in the **stance the preset expects** (e.g. quadruped with all fours planted). Extra creature motion beyond the single locomotion preset should be procedural, a second rig, or external clips.
+
+---
+
+## Quick map
+
+| `rig_type` | Preset(s) |
+| --- | --- |
+| `biped` | For v1 use `preset:biped:*` below; v2.5 uses its separate catalog above. Only legacy comma-string/default mode maps plain names. |
+| `quadruped` | `preset:quadruped:walk` |
+| `hexapod` | `preset:hexapod:walk` |
+| `octopod` | `preset:octopod:walk` |
+| `serpentine` | `preset:serpentine:march` |
+| `aquatic` | `preset:aquatic:march` |
+| `avian` | (none catalogued — procedural / external) |
+
+---
+
+## Creature locomotion
+
+### Quadruped (`rig_type=quadruped`)
+
+```text
+preset:quadruped:walk
+```
+
+### Hexapod (`rig_type=hexapod`)
+
+```text
+preset:hexapod:walk
+```
+
+### Octopod (`rig_type=octopod`)
+
+```text
+preset:octopod:walk
+```
+
+### Serpentine (`rig_type=serpentine`)
+
+```text
+preset:serpentine:march
+```
+
+### Aquatic (`rig_type=aquatic`)
+
+```text
+preset:aquatic:march
+```
+
+### Avian (`rig_type=avian`)
+
+No dedicated Tripo locomotion preset in this catalog. Prefer procedural wing/idle motion, a ground `rig_type` for walk when applicable, or external clips.
+
+---
+
+## Biped (`rig_type=biped`, `preset:biped:*`)
+
+Prefer the full `preset:biped:...` form. In legacy comma-string mode, plain `preset:idle` is mapped server-side to `preset:biped:idle` (and similar).
+
+### Locomotion / core gameplay (most common)
+
+```text
+preset:biped:idle
+preset:biped:walk
+preset:biped:run
+preset:biped:jump
+preset:biped:jump_down
+preset:biped:turn
+preset:biped:climb
+preset:biped:fall
+preset:biped:flee_01
+preset:biped:flee_02
+preset:biped:swim
+preset:biped:run_upstairs
+preset:biped:wait
+preset:biped:standing_relax
+preset:biped:swagger
+```
+
+### Combat / action
+
+```text
+preset:biped:slash
+preset:biped:shoot
+preset:biped:cast_a_spell
+preset:biped:fire
+preset:biped:chop
+preset:biped:front_kick_01
+preset:biped:front_kick_02
+preset:biped:box_01
+preset:biped:box_02
+preset:biped:box_03
+preset:biped:hit_to_body_01
+preset:biped:hit_to_body_02
+preset:biped:hit_to_head
+preset:biped:hit_to_side
+preset:biped:hit_to_stomach
+preset:biped:hurt
+preset:biped:dive
+preset:biped:flip
+```
+
+### Social / emote
+
+```text
+preset:biped:agree
+preset:biped:bow
+preset:biped:cheer
+preset:biped:clap
+preset:biped:greet_01
+preset:biped:greet_02
+preset:biped:greet_03
+preset:biped:greet_04
+preset:biped:wave_goodbye_01
+preset:biped:wave_goodbye_02
+preset:biped:hug
+preset:biped:fold_arms
+preset:biped:look_around
+preset:biped:scratch
+preset:biped:heart_pose
+```
+
+### Emotion
+
+```text
+preset:biped:afraid
+preset:biped:angry_01
+preset:biped:angry_02
+preset:biped:angry_03
+preset:biped:cry
+preset:biped:sob
+preset:biped:depressed
+preset:biped:frightened
+preset:biped:frustrated_01
+preset:biped:frustrated_02
+preset:biped:scared_01
+preset:biped:scared_02
+preset:biped:laugh_01
+preset:biped:laugh_02
+preset:biped:complain_01
+preset:biped:complain_02
+preset:biped:defeat_02
+preset:biped:defeat_03
+preset:biped:victory_celebration
+preset:biped:freaky
+```
+
+### Dance / performance
+
+```text
+preset:biped:dance_01
+preset:biped:dance_02
+preset:biped:dance_03
+preset:biped:dance_04
+preset:biped:dance_05
+preset:biped:dance_06
+preset:biped:sing_01
+preset:biped:sing_02
+preset:biped:sing_03
+preset:biped:sing_04
+```
+
+### Sports / activity
+
+```text
+preset:biped:basketball_shot
+preset:biped:crossover_dribble
+preset:biped:dribble
+preset:biped:football_catch
+preset:biped:football_save
+preset:biped:football_pass
+preset:biped:golf
+preset:biped:pitch_baseball
+preset:biped:volleyball
+preset:biped:surf
+preset:biped:jump_rope_01
+preset:biped:jump_rope_02
+preset:biped:cross_body_crunch
+preset:biped:press-up
+preset:biped:warm_up
+preset:biped:lift_heavy
+```
+
+### Work / props / misc
+
+```text
+preset:biped:dig
+preset:biped:shovel
+preset:biped:sit
+preset:biped:make_a_call_01
+preset:biped:make_a_call_02
+preset:biped:play_mobile_game
+preset:biped:play_video_game
+```
+
+### Biped full flat list (copy-paste)
+
+```text
+preset:biped:afraid
+preset:biped:agree
+preset:biped:angry_01
+preset:biped:angry_02
+preset:biped:angry_03
+preset:biped:basketball_shot
+preset:biped:bow
+preset:biped:box_01
+preset:biped:box_02
+preset:biped:box_03
+preset:biped:cast_a_spell
+preset:biped:cheer
+preset:biped:chop
+preset:biped:clap
+preset:biped:climb
+preset:biped:complain_01
+preset:biped:complain_02
+preset:biped:cross_body_crunch
+preset:biped:crossover_dribble
+preset:biped:cry
+preset:biped:dance_01
+preset:biped:dance_02
+preset:biped:dance_03
+preset:biped:dance_04
+preset:biped:dance_05
+preset:biped:dance_06
+preset:biped:defeat_02
+preset:biped:defeat_03
+preset:biped:depressed
+preset:biped:dig
+preset:biped:dive
+preset:biped:dribble
+preset:biped:fall
+preset:biped:fire
+preset:biped:flee_01
+preset:biped:flee_02
+preset:biped:flip
+preset:biped:fold_arms
+preset:biped:football_catch
+preset:biped:football_save
+preset:biped:football_pass
+preset:biped:freaky
+preset:biped:frightened
+preset:biped:front_kick_01
+preset:biped:front_kick_02
+preset:biped:frustrated_01
+preset:biped:frustrated_02
+preset:biped:golf
+preset:biped:greet_01
+preset:biped:greet_02
+preset:biped:greet_03
+preset:biped:greet_04
+preset:biped:heart_pose
+preset:biped:hit_to_body_01
+preset:biped:hit_to_body_02
+preset:biped:hit_to_head
+preset:biped:hit_to_side
+preset:biped:hit_to_stomach
+preset:biped:hug
+preset:biped:hurt
+preset:biped:idle
+preset:biped:jump_down
+preset:biped:jump
+preset:biped:jump_rope_01
+preset:biped:jump_rope_02
+preset:biped:laugh_01
+preset:biped:laugh_02
+preset:biped:lift_heavy
+preset:biped:look_around
+preset:biped:make_a_call_01
+preset:biped:make_a_call_02
+preset:biped:pitch_baseball
+preset:biped:play_mobile_game
+preset:biped:play_video_game
+preset:biped:press-up
+preset:biped:run_upstairs
+preset:biped:run
+preset:biped:scared_01
+preset:biped:scared_02
+preset:biped:scratch
+preset:biped:shoot
+preset:biped:shovel
+preset:biped:sing_01
+preset:biped:sing_02
+preset:biped:sing_03
+preset:biped:sing_04
+preset:biped:sit
+preset:biped:slash
+preset:biped:sob
+preset:biped:standing_relax
+preset:biped:surf
+preset:biped:swagger
+preset:biped:swim
+preset:biped:turn
+preset:biped:victory_celebration
+preset:biped:volleyball
+preset:biped:wait
+preset:biped:walk
+preset:biped:warm_up
+preset:biped:wave_goodbye_01
+preset:biped:wave_goodbye_02
+```