lopatin/snaketron/author-skin
Claude Code and agent skill by lopatin/snaketron
- community
Checking that the SKILL.md file is still in the repository.
Turn an exact approved prototype or explicitly selected draft candidate into a validated private SkinDoc v2 implementation using layers, textures, sprite sheets, reusable modifiers, or a hybrid. Use for factory builds and capability-gated interactive draft submissions; do not use for prototype approval or publication.
Install
npx vskill@latest install lopatin/snaketron/author-skinSKILL.md
Author a Snaketron skin
Build one exact retained prototype candidate as a SkinDoc v2 revision. This package is agent-neutral: use the inputs and output schemas supplied by the caller, not an agent's transcript, private memory, or provider-specific executable.
Required inputs
Stop with a typed invalid_input result if any item is absent or inconsistent:
- prototype bytes or a readable content-addressed reference;
- its prototype manifest and one exact
input_authorityrecord in either
approved_prototype or draft_submission mode;
- the repository-canonical design direction and literal human feedback;
- the generated capability manifest for the pinned schema/renderer SHA and a
separate literal host-operation capability record (authority modes, operations, and any tighter driver limits);
- output schemas, budgets, pure-tool allowlist, and artifact workspace.
The authority artifact hash must equal the prototype image hash. In approved_prototype, require the exact human prototype_approval. In draft_submission, require the agent-selected candidate hash, literal selection rationale, and retained selection-record hash; never describe that record as human approval. Never substitute another candidate or silently regenerate the selected direction.
Load the contract
Read references/contract.md completely. It is locked policy. Always read the locked shared Skin Design Guidelines completely and apply them in the implementation plan. Then read only the other craft references needed for this build:
- Always read references/prototypes.md and
- For procedural layers or effects, read
- For any image source, read
references/textures-sprites.md.
- For separable components, reusable modifier objects, extraction, or video,
also read references/modifiers-video.md.
- Before returning a result, read
Author in two logical passes
- Inspect the prototype at game scale and identify the few visual features
that carry its identity. Treat its implementation_hint as advice.
- Choose
layers,texture,sprite_sheet, orhybridusing the routing
rules in the playbook. Prefer layers for patterns and formula-driven motion.
- Before requesting or generating assets, emit
implementation-plan.json
conforming to schemas/implementation-plan.schema.json. Fill its bounded design_guidelines object from the shared guidelines; these decisions must agree with the executable layer, asset, wrap, and animation plan. Record input_authority, the common animation period, and every image asset's natural X length and derived frame count Y, plus placement and wrap axes. Map every asset one-to-one to a modifier record, image layer, and earlier fallback. Put the reason for changing the prototype hint in rationale.
- Compose the v2 document draft. For each planned image asset, declare one
matching texture name and use the same asset_index and texture_name in its request. Put a complete ordinary fallback immediately below every image layer. The only unresolved reference is pending:asset:<index>, with the planned kind and no descriptor. Do not fabricate a generated hash, descriptor, upload, or successful gate.
- If ordinary generated assets are needed, emit
generate_assettool
requests whose arguments conform to schemas/asset-request.schema.json for the caller's image_generator role. Emit edit_asset only in a repair phase the caller explicitly advertises; otherwise the deterministic forge owns its isolated local LaMa helper. The configured generator default may be Gemini 3 Pro Image, but the skill never requires Gemini or calls a provider directly when running as a factory task worker. Extraction, modifier reuse, and video are separate journaled host capabilities: use their request schemas only when the caller advertises the exact operation and retained-input/output contract. Otherwise return platform_gap.
- Return only the plan, v2-shaped document draft, tool requests, trace, usage,
or typed failure allowed by the caller's worker-result schema. If the prototype cannot be expressed by the pinned capabilities, return platform_gap; do not create a Rust skin.
- The driver journals every external operation, generates/repairs assets,
binds exact served refs and descriptors into the draft, validates the exact bytes, registers the private revision, and captures real-browser evidence. A later bounded repair request may supply failed artifacts and reports, but neither worker nor driver overwrites their history. A pure factory worker never uploads, registers, approves, publishes, uses Git, or mutates external storage, and the factory accepts only approved_prototype. An explicitly authorized interactive agent acting as the advertised host may journal and execute the declared generation/extraction/binding operations, register the exact private revision, and request Snaketron admin review. It may never publish; an unadvertised operation is platform_gap.
The factory explicitly bundles this directory and records its SHA. Agent discovery wrappers are conveniences only; see references/integration.md.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/author-skin/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/lopatin/snaketron/author-skin)