SkillSmith

Changelog

New updates and improvements to SkillSmith.

v5.1.0
  • NewDashboard launchpad — your home base after login with a prominent "Create a New Skill" hero, your recent skills, saved favorites, and quick stats.
  • NewMy Skills library — browse, edit, download, or delete all skills you've created. Edit re-opens the wizard with your skill pre-loaded.
  • NewFavorites — save skills from the gallery with a heart icon. Find them in your Saved tab. Works on gallery cards and detail pages.
  • ImprovedDashboard is now server-rendered for instant load. Settings moved to a dedicated tab.
v5.0.0
  • NewGoogle OAuth login — sign in with your Google account to unlock API tokens and the user dashboard. The skill creation wizard remains fully public.
  • NewPublic REST API v1 — 10 versioned endpoints for programmatic access to all AI pipelines. Authenticate with Bearer tokens (sk_<hash>).
  • NewPer-user API tokens with labels and read/write access. SHA-256 hashed storage. Free: 3 tokens / 30 req/hr. Premium: 10 tokens / 100 req/hr.
  • NewUser dashboard — manage your account, tier, and API tokens at /dashboard.
  • ImprovedAtomic Redis rate limiting (fixed INCR/EXPIRE race condition). Per-user limits for API v1.
v4.0.0
  • NewEngine Settings page — admin-only view showing full architecture, AI pipeline flow, all 6 master prompts, and gallery moderation pipeline
  • ImprovedAI prompts updated with 2026 skills ecosystem knowledge — open standard patterns, multi-agent coordination (OpenClaw), self-improving agents, solo founder cron-agent pattern
  • New2 new skill categories (Agent Orchestration, Self-Improvement) + 5 design patterns (Cron Agent, Coordination Skill, Knowledge Encoder, Failure-to-Skill Loop, MCP Bridge)
  • ImprovedQuality analyzer now evaluates Domain Value and Composability — skills that just repackage built-in capabilities score lower
v3.5.2
  • FixedGallery submissions broken (500 error) — schema drift caused queries to reference a missing database column. All queries now use explicit column lists instead of bare .returning()/.select()
  • FixedSlug collision check now wrapped in try/catch for proper error handling
v3.5.1
  • FixedAll AI generation routes were broken (500 error) due to deprecated model ID. Updated from claude-sonnet-4-20250514 to claude-sonnet-4-6 (Sonnet 4.6)
  • ImprovedError messages now surface actual API error details for easier debugging
v3.5.0
  • NewSkill testing sandbox — new Test tab simulates how Claude Code responds with your skill loaded, with streaming output and 5 tests per session
  • NewStructured iteration loop — Analyze Quality scores your skill 1-10 with strengths, weaknesses, and clickable suggestion chips that auto-fill the refine prompt
  • NewBefore/after diff view — Changes tab shows GitHub-style green/red diff after refinements, with one-click revert to previous version
  • ImprovedLoading skeletons — gallery, admin dashboard, and skill generation now show skeleton layouts instead of blank screens during loading
  • ImprovedRefine flow now saves previous skill state automatically, enabling diff view and revert
v3.0.0
  • NewMulti-file skill generation — skills now generate as complete packages with SKILL.md plus companion files in scripts/, references/, examples/, and assets/ folders
  • NewAdvanced frontmatter configurator — two-way bound sidebar UI for all official Claude Code skill fields: argument-hint, allowed-tools, context, agent, model, and more
  • NewDescription optimization tool — AI generates 10 eval queries (5 should-trigger, 5 near-miss should-not), scores your description accuracy, and suggests improvements
  • NewComplexity tiers — Simple, Moderate, and Complex badges on skill ideas control how many companion files are generated. Click to override before generating.
  • NewDeep profile parser — paste your LinkedIn, GitHub, portfolio, booking page, or resume and the system extracts every signal to suggest eerily specific skills
  • NewDark mode — system, light, and dark themes with toggle in the navbar
  • NewError boundaries — friendly crash recovery with Start Over and Try Again options
  • ImprovedGenerate API switched from streaming text to structured JSON output (generateObject + Zod) for reliable multi-file generation
  • ImprovedAll AI prompts rewritten with progressive disclosure enforcement and complexity-tier-specific instructions
  • InfraAdded companion_files jsonb column to database schema for gallery skills
v2.0.0
  • NewPublic gallery — browse, search, and download community-created skills at /gallery
  • NewShare to Gallery — submit skills with optional author attribution, reviewed by AI before publishing
  • NewAI auto-review — every submission scored for safety, quality, usefulness, and originality by Claude Sonnet 4
  • NewAdmin dashboard at /admin — moderation queue, stats cards, approve/reject/re-review actions
  • NewIndividual skill pages with rendered preview, metadata sidebar, and download button
  • ImprovedRebrand: "Skill Maker" → "SkillSmith" across all UI and metadata
  • ImprovedPersona icons now use clean lucide-react SVGs instead of emojis
  • InfraSwitched database from Neon HTTP to Railway Postgres (postgres.js driver)
  • InfraAdded Redis-based rate limiting (ioredis) with in-memory fallback
  • InfraNew shared_skills table with moderation status flow and AI review scores
v1.5.0
  • NewImage paste & upload — paste screenshots (Cmd+V) anywhere, or drag & drop images. Claude reads them directly via vision API.
  • New6 example personas (Frontend Dev, Data Engineer, Mobile Dev, Python Backend, Product Manager, DevOps/SRE) — click to preview, import into the prompt box
  • NewDynamic OG image — branded card with feature pills, generated server-side via Next.js ImageResponse
  • ImprovedUpload zone now accepts images (PNG, JPG, WebP, GIF) up to 500KB, with thumbnail previews and hover-to-remove
  • ImprovedMax file limit increased from 5 to 10 (images + text combined)
v1.4.0
  • NewMajor prompt rewrite — all 3 AI prompts now use the complete 2026 Claude Code skills spec
  • ImprovedGenerated skills now use advanced features: argument-hint, allowed-tools, dynamic context injection (!`command`), context: fork, skill folder structure
  • ImprovedBetter few-shot examples using Anthropic's 3 skill categories (Document Creation, Workflow Automation, MCP Enhancement)
  • ImprovedProgressive disclosure guidance — skills reference linked files instead of embedding everything inline
  • FixedOpus 4.6 model ID corrected — was returning empty responses due to wrong version date
  • FixedAPI routes now return proper 500 errors instead of silently empty streams on failure
v1.3.0
  • NewAdmin panel — full system transparency dialog showing architecture, API config, validation rules, database schema, and exact prompts
  • ImprovedAdmin modal expanded to max-w-6xl with proper top positioning (no title clipping)
v1.2.0
  • NewModel selector — Sonnet 4 (free), Opus 4.5 and Opus 4.6 (unlock with code)
  • NewTwo-tier unlock system: premium code (models only) and admin code (models + prompt transparency)
  • NewPrompt viewer tab — admin users can see exact system/user prompts sent to Anthropic
  • ImprovedAll API routes validate model + unlock code server-side on every request
v1.1.0
  • NewNeon Postgres database for analytics and skill archiving
  • NewFire-and-forget analytics tracking on every AI generation
  • NewSave-to-database on skill download
  • NewPublic changelog page
  • ImprovedLanding page redesign — "How it's built" section, skill explainer, tech stack badges
  • ImprovedFooter: built-by attribution, skill docs link, changelog link
v1.0.0
  • New4-step creation wizard: Context → Ideas → Draft → Download
  • NewAI-powered context analysis — suggests 2-3 tailored skill ideas
  • NewFull SKILL.md generation with streaming output via Claude Sonnet 4
  • NewAI refinement — describe changes in natural language
  • NewReal-time validation: YAML frontmatter, kebab-case name, description length
  • NewSplit-pane editor with live Markdown preview
  • NewClient-side .zip download via JSZip
  • NewFile upload support (.md, .txt, .json, .yaml, .py, .ts, .js)
  • NewlocalStorage persistence — resume where you left off
  • NewIn-memory rate limiting (10 req/min per IP)
  • NewLanding page with hero, how-it-works, and CTAs
  • InfraNext.js 16, React 19, Tailwind CSS, Vercel AI SDK v6, Railway deployment
  • Infra26 tests covering validator, format parser, and zip generator