aozu.mantle.tools
AOZU is an agent-native workshop for creating, aligning, and exporting reusable layered character assets with WebMCP.
First seen 2026-09-21 · Last seen 2026-09-28 · Source: webmcp.com
Answer 0 · Act 16 · Transact 0 · Score: Not yet scored
Tool inspector
replace_character_asset
Install one complete canonical Character layer after inspect_character_contract with scope:appearance. The first Appearance body establishes a front A-pose: upright torso, neutral face, arms angled down and away from the torso, relaxed visible hands, stable feet, and the complete head-to-feet silhouette inside the canvas. This is the AOZU authoring convention, not a claim of skeleton compatibility. Preserve this pose and registration in derived outfits and expressions. Existing artwork is not automatically verified A-pose; review it visually before proposing a baseline change, which can make derived layers stale. Follow its backgroundPreparation workflow: solid-color generation, removal with a permitted environment tool, then alpha/edge verification. This is a true replacement without preserving old pixels. Expressions contain only a complete whole head with transparency elsewhere. Outfits contain the complete dressed character skin compatible with the reference pose; exact base-pixel coverage is not required. Opaque input is rejected; AOZU never removes backgrounds. Rejected or stale input does not mutate or navigate. Submit exact 512×768 RGBA or explicitly request the inspected normalization. After variant acceptance, follow the returned alignment.visualReview through all four browser modes before the next asset; review the canonical body in its regular Composite preview. AOZU itself handles effects.navigation in this browser tab; the agent must not repeat that navigation. Observe the resulting page after rendering, then call inspect_workspace again for fresh context. A successful tool result is not proof of visual correctness.
Input schema
| Property | Type | Required | Description |
|---|---|---|---|
characterId |
string | yes | — |
dataUrl |
string | yes | — |
expectedAssetSha256 |
string,null | yes | — |
expectedRevision |
integer | yes | — |
filename |
string | yes | — |
group |
— | yes | — |
label |
string | yes | — |
layer |
— | yes | — |
normalization |
object | no | — |
variantId |
string | yes | — |
Raw JSON schema
{
"additionalProperties": false,
"properties": {
"characterId": {
"minLength": 1,
"type": "string"
},
"dataUrl": {
"maxLength": 7100000,
"pattern": "^data:image/png;base64,",
"type": "string"
},
"expectedAssetSha256": {
"pattern": "^[0-9a-f]{64}$",
"type": [
"string",
"null"
]
},
"expectedRevision": {
"minimum": 0,
"type": "integer"
},
"filename": {
"maxLength": 200,
"minLength": 1,
"type": "string"
},
"group": {
"enum": [
"body",
"expression",
"outfit",
"prop"
]
},
"label": {
"maxLength": 80,
"minLength": 1,
"type": "string"
},
"layer": {
"enum": [
"body",
"head",
"back",
"front"
]
},
"normalization": {
"additionalProperties": false,
"properties": {
"align": {
"enum": [
"none",
"reference-visible-bounds"
]
},
"resize": {
"enum": [
"none",
"exact-aspect-downscale"
]
}
},
"required": [
"resize",
"align"
],
"type": "object"
},
"variantId": {
"pattern": "^[a-z0-9][a-z0-9_-]{0,39}$",
"type": "string"
}
},
"required": [
"characterId",
"group",
"variantId",
"label",
"layer",
"expectedRevision",
"expectedAssetSha256",
"filename",
"dataUrl"
],
"type": "object"
}
Similar websites
tracinginsights.com
Analyze Formula 1 racing data with interactive charts for lap times, live telemetry, timing data and race performance. Explore and Compare drivers across multip…
rough-cut.samarthsaxena1672003.workers.dev
A human-first video editor with structured WebMCP tools for any compatible agent.
fiveminutebrief.ai
panoptik-studio.vercel.app
Browser-native demo video editor where humans and AI agents co-edit on the same canvas via WebMCP. No uploads, no server.