Skip to content

RandallLiuXin/GodotMaker / skills/core/gm-asset

gm-asset

Verified

Asset stage manager. Reads current-tag ASSETS.md gaps, accepts user-provided assets, plans visual production units, dispatches asset-producer subagents, collects validated Asset Skill results, updates ASSETS.md directly, and emits minimal worker runtime snapshots from that one authority. Explicit invocation only - use /gm-asset.

references/providers/wan.md

View on GitHub
# Wan 2.7 Image Provider

Use this file when `.godotmaker/config.yaml` sets `asset_image_model` to `wan`
or `wan:<model>`. The bare selector means `wan:wan2.7-image`; use
`wan:wan2.7-image-pro` explicitly for the higher-quality model. Do not use a
`wanxiang` alias.

## Source Generation

Write one spec per generated source under `.godotmaker/asset-generation/specs/`:

```json
{
  "asset_id": "<asset_id>",
  "model": "wan:wan2.7-image",
  "prompt": "<full prompt>",
  "prompt_path": ".godotmaker/asset-generation/prompts/<asset_id>.txt",
  "source_path": ".godotmaker/asset-generation/sources/<asset_id>_source.png",
  "size": "1K",
  "aspect_ratio": "16:9",
  "reference_inputs": [],
  "report_path": ".godotmaker/asset-generation/reports/<asset_id>_source.json"
}
```

Run:

```bash
python tools/asset_source_generate.py --spec <spec.json>
```

The tool calls Wan synchronously, downloads the short-lived returned URL in the
same execution, verifies it is PNG, and materializes the exact `source_path`.
The report records request ID, model, operation, usage, dimensions, and
role/path/hash provenance, but never an API key, Base64 image data, or result
URL.

## Requirements and regional routing

1. Set `DASHSCOPE_API_KEY`.
2. Set `DASHSCOPE_REGION` explicitly to `beijing` or `singapore`. API keys are
   regional; the tool never infers a region from a key.
3. Optionally set `DASHSCOPE_BASE_URL` to the matching public endpoint or a
   matching business-space endpoint. Use either the host or its `/api/v1`
   base (with or without a trailing slash); the full Wan generation endpoint
   is also accepted. Defaults are
   `https://dashscope.aliyuncs.com` for Beijing and
   `https://dashscope-intl.aliyuncs.com` for Singapore.
4. Run `python tools/check_env.py` after configuring the provider.

The preferred business-space forms are
`https://<workspace-id>.cn-beijing.maas.aliyuncs.com` and
`https://<workspace-id>.ap-southeast-1.maas.aliyuncs.com`. A region/base URL
mismatch fails before a network request.

## References and limits

Wan accepts 0–9 images. Put every required local image in `reference_inputs` as
`{ "role": "canonical|style", "path": "<local image path>" }` in the
planned order. The generator sends each actual file as a Base64 data URI in that
same order, then the prompt. It validates up to 20 MB per image before Base64
encoding, a total encoded reference payload of at most 32 MB, 240–8000 px per
edge, a 1:8–8:1 ratio, and supported JPEG/JPG/PNG/BMP/WEBP formats.

Wan 2.7 does not accept transparent PNG inputs. This initial integration fails
closed only when actual alpha values are present; an RGBA PNG whose alpha is
fully opaque is accepted. Provide an opaque image rather than silently
flattening it. `wan2.7-image-pro` permits 4K only for text-to-image;
image editing is limited to 2K. The generator maps the shared `size` and
`aspect_ratio` deterministically to Wan's `WIDTH*HEIGHT` parameter; accepted
sizes are `1K`, `2K`, and `4K` (`512` is rejected), and it always
sends `watermark: false`.

## Handoff

After generation, keep the controlled source report with the processing and
finalization reports. Do not reuse a remote result URL as an asset, write
provenance by hand, substitute another provider, or retry through a different
provider.

## Explicit live smoke

Normal tests mock every Wan request. After configuring a non-production test
workspace and regional credentials, run the two text/reference paths twice each:

```bash
WAN_LIVE_SMOKE=1 python -m pytest tests/tools/test_wan_live_smoke.py -q
```

The fixed seed makes outputs comparatively stable but does not assert identical
pixels. Preserve the generated source and report for every successful request.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall