MohamedAbdallah-14/prompt-to-asset

Plugin skill set rounded out from 8 asset-type skills to 12, closing long-tail gaps in SVG authoring, prompt-dialect rewriting, validation repair, and multi-asset brand consistency.

Added (4 new skills)

• svg-authoring — engaged when asset_generate_* returns an InlineSvgPlan. Enforces viewBox, path budget, palette, optical balance, and small-scale (16×16) legibility so the emitted <svg> passes asset_save_inline_svg validation. Covers style taxonomy (flat / outlined / filled / duotone / minimal / brutalist) with concrete SVG patterns for each.
• t2i-prompt-dialect — engaged during prompt rewriting. Per-model rules for gpt-image-1, Imagen / Gemini, SD 1.5 / SDXL, Flux.1 / Flux.2, Midjourney, Ideogram, Recraft. Handles negative-prompt translation (Flux errors on it, Imagen / gpt-image-1 ignore it, SDXL uses it), token budgets (SDXL 77-token CLIP limit with BREAK chunking), transparency quirks (never prompt Imagen / Gemini for transparency), and brand-palette injection per dialect.
• asset-validation-debug — engaged when asset_validate or a generator returns warnings. Maps failure codes (T0_CHECKERBOARD, T0_ALPHA_MISSING, T1_PALETTE_DRIFT, T1_TEXT_MISSPELL, etc.) to concrete repair primitives (matte, inpaint, route change, seed sweep, composite SVG type) with cost ratios (SVGO 0.01×, matte 0.05×, inpaint 0.3×, regenerate 4×). Applies a retry budget so Claude does not loop on hopeless regenerations.
• brand-consistency — engaged for multi-asset sets. Builds BrandBundle (palette, typography, style refs, do-not list, logo mark), enforces palette per model (Recraft controls.colors hard-lock → IP-Adapter → Midjourney --sref → Flux LoRA), validates CSD style similarity + ΔE2000, promotes accepted assets back into style_refs[] so each new generation tightens the brand lock. Covers LoRA-training ROI (break-even ≈ 20 assets).

Unchanged

• MCP tool surface — still 24 tools. No API changes.
• mcp-server npm package — still 0.3.0 (no code changes). This release tags the plugin skill bundle only.
• Existing 8 skills — asset-enhancer, logo, app-icon, favicon, og-image, illustration, transparent-bg, vectorize. Unchanged.

Mirror system

scripts/sync-mirrors.sh and scripts/verify-repo.sh propagate every SKILL.md to .claude/, .cursor/, .windsurf/, and plugins/prompt-to-asset/ under byte-for-byte CI enforcement. rules/asset-enhancer-activate.md gained a "Supporting skills" section that regenerates into CLAUDE.md, AGENTS.md, GEMINI.md, .clinerules/, .cursor/rules/, .windsurf/rules/, and .github/copilot-instructions.md.

Research

Full design specs under docs/research/24-skills-for-p2a/ — 9 documents covering gap analysis, marketplace-skill survey, per-skill design, and the master SYNTHESIS.md with dependency map and ranked priority matrix.

CI

All 8 CI jobs green across ubuntu/macos/windows × Node 20/22/24. No test changes; every existing test still passes.