v1.17.0 — Chrome Extension & MCP Server Fix

Chrome Extension — Save as Skill (#60, #62)

Save any webpage as a SKILL.md with one click — fully browser-based, no server needed.

• Popup UI: Click extension icon → save current page as a skill
• Context menu: Right-click → "Save page as Skill" / "Save selection as Skill"
• Auto-tags: Smart detection from 40+ tech keywords
• Markdown conversion: HTML → clean markdown via Turndown
• Downloads API: Saves to Downloads/skillkit-skills/ — zero server dependency

Website & Docs

• "Add to Chrome" button in Hero section
• "Extension" nav link
• New chrome-extension.mdx docs page
• README section with install instructions

MCP Server Fix (#58, #63)

Fixed @skillkit/mcp silently exiting when invoked via npx:

• Symlink resolution: isMain guard now uses realpathSync() on both process.argv[1] and import.meta.url
• Graceful fallback: try/catch prevents crashes when imported as a library
• skills.json bundled: prebuild copies marketplace/skills.json into package data/ dir (15,065 skills)
• Version string: Updated from 1.12.0 to 1.17.0

POST /save API Endpoint

New endpoint for CLI/programmatic skill saving with SSRF protection:

• IPv6 bracket handling, IPv6-mapped IPv4, full loopback range
• Multicast/Class E blocking
• DOMException-based timeout detection

v1.16.0 - 44 Agents Support

What's New

12 New AI Agents (32 → 44 total)

Added support for: Devin, Aider, Sourcegraph Cody, Amazon Q Developer, Augment Code, Replit Agent, Bolt, Lovable, Tabby, Tabnine, CodeGPT, PlayCode Agent

Each agent gets proper config (skillsDir, configFile, format) and auto-detection.

GenericAgentAdapter (Bug Fix)

Fixed critical bug where 26 agents shared UniversalAdapter, causing wrong type, skillsDir, and name. Replaced with GenericAgentAdapter that reads per-agent config from AGENT_CONFIG.

Updated Banner

Refreshed monochromatic SVG banner with updated agent count and improved spacing.

Full Changelog

• feat: add 12 missing AI coding agents
• fix: replace UniversalAdapter with GenericAgentAdapter for all non-universal agents
• fix: expand translateSkillToAll to include all 44 agents
• test: 115+ adapter tests for GenericAgentAdapter correctness
• style: updated banner and README for 44 agents
• docs: updated all agent counts across documentation

v1.15.0

What's New

Security Scanner (#50)

• 46+ detection rules across 6 threat categories: prompt injection, command injection, data exfiltration, tool abuse, hardcoded secrets, and unicode steganography
• 3 analyzers: StaticAnalyzer (regex pattern matching), ManifestAnalyzer (SKILL.md validation), SecretsAnalyzer (credential detection)
• CLI command: skillkit scan <path> with --format (summary/json/table/sarif), --fail-on, --skip-rules
• Install integration: Skills are automatically scanned during skillkit install (use --no-scan to skip)
• Publish integration: Auto-scan before skillkit publish
• SARIF output: Upload to GitHub Code Scanning for IDE integration

TUI Security Scan Screen (#53)

• New Scan screen in the terminal UI (press z to navigate)
• Live scan execution with severity-colored findings
• Finding navigation with j/k keys, rescan with r
• Windowed list rendering for large result sets

Security Documentation (#53)

• New dedicated Security Scanning docs page with full rule reference
• All 46+ rules documented with IDs, descriptions, and severity levels
• CI/CD integration guide with GitHub Actions example
• Programmatic SkillScanner API documentation

Website Fixes

• Fix initial stars loader (#51): Removed stale hardcoded values, shows dash placeholder while loading
• Fix compatibility matrix (#52): Rewrote with accurate agent capability data from real agent adapters
• Product Hunt top post badge (#54): Updated from "featured" to "top post daily" badge
• Live PH upvote count (#49): Embed badge with live count

Commits since v1.14.0

• feat: add security scanner for detecting malicious patterns in skills (#50)
• fix: remove hardcoded initial stars count from website loader (#51)
• fix: rewrite compatibility matrix with accurate agent capability data (#52)
• docs: add security scanner documentation, TUI screen, and README updates (#53)
• feat: update Product Hunt badge to top post daily badge (#54)
• fix: use PH embed badge image with live upvote count (#49)
• feat: show live Product Hunt upvotes, remove Apache license from stats bar (#48)
• feat: add Product Hunt featured badge to website stats bar (#47)
• feat: add live GitHub stars CTA and redesign README banner (#46)

Full Changelog: v1.14.0...v1.15.0

What's Changed

• feat: add live GitHub stars CTA and redesign README banner by @rohitg00 in #46
• feat: add Product Hunt badge to website by @rohitg00 in #47
• feat: live PH upvotes in stats bar, remove license badge by @rohitg00 in #48
• fix: use PH embed badge with live upvotes by @rohitg00 in #49
• feat: add security scanner for detecting malicious patterns in skills by @rohitg00 in #50
• fix: remove hardcoded initial stars count from website by @rohitg00 in #51
• fix: rewrite compatibility matrix with accurate agent data by @rohitg00 in #52
• feat: update Product Hunt badge to top post daily by @rohitg00 in #54
• docs: add security scanner docs, TUI screen, and README updates by @rohitg00 in #53

Full Changelog: v1.14.0...v1.15.0

v1.14.0 - Smart Context Generation

🚀 Smart Context Generation

SkillKit now includes AI-powered skill generation that goes beyond simple LLM prompts. The new skillkit generate command pulls context from 4 sources and creates agent-optimized skills.

✨ Key Features

Multi-Provider LLM Support

• Claude (Anthropic)
• GPT-4/GPT-4o (OpenAI)
• Gemini (Google)
• Ollama (local models)
• OpenRouter (100+ models)

4 Context Sources

• 📚 Documentation via Context7 MCP
• 💻 Your codebase patterns
• 🛒 15,000+ marketplace skills
• 🧠 Your corrections and learnings

Agent-Specific Optimization

• Generates tailored output for 32 agents
• Trust scoring (0-10) with grades
• Compatibility matrix per agent

Skill Composition

skillkit generate --compose "testing patterns for react"

Natural language search finds and composes existing skills automatically.

📖 Usage

# Interactive wizard (recommended)
skillkit generate

# Use specific provider
skillkit generate --provider openai --model gpt-4o

# Target specific agents
skillkit generate --agents claude-code,cursor

🔧 Installation

npx skillkit@latest

📋 Full Changelog

New Files (40+)

• packages/core/src/ai/providers/ - 6 LLM providers
• packages/core/src/ai/context/ - 4 context sources
• packages/core/src/ai/composition/ - Skill composition engine
• packages/core/src/ai/agents/ - Agent optimization
• packages/core/src/ai/wizard/ - Interactive wizard
• packages/core/src/ai/security/ - Trust scoring + injection detection
• packages/cli/src/commands/generate.ts - Generate command

Documentation

• Updated CLI commands reference
• Added Generate Command section
• Updated website Advanced Features

---

Full documentation

v1.13.0 - Memory Lifecycle Hooks & Progressive Disclosure

What's New

Memory Lifecycle Hooks

Automatic memory capture and injection via Claude Code lifecycle hooks:

Hook	Trigger	Action
SessionStart	Session begins	Inject relevant learnings
PostToolUse	Tool completes	Capture outcomes as observations
SessionEnd	Session closes	Compress observations to learnings

Progressive Disclosure (Token Optimization)

3-layer retrieval system for efficient context usage:

Layer	Content	~Tokens
Index	Titles, tags, timestamps	50-100
Timeline	Context, excerpts, activity	~200
Details	Full content, metadata	500-1000

Auto-CLAUDE.md Updates

Sync your most effective learnings to CLAUDE.md:

skillkit memory sync-claude

This populates the ## LEARNED section with high-effectiveness insights.

New CLI Commands

skillkit memory sync-claude   # Update CLAUDE.md with learnings
skillkit memory index         # View memory index (Layer 1)

API Additions

MemoryHookManager

import { MemoryHookManager } from '@skillkit/core'

const manager = new MemoryHookManager(projectPath)
await manager.onSessionStart()
await manager.onToolUse({ tool_name: 'Write', ... })
await manager.onSessionEnd()

ProgressiveDisclosureManager

import { ProgressiveDisclosureManager } from '@skillkit/core'

const pdm = new ProgressiveDisclosureManager(projectPath)
const index = pdm.getIndex({ includeGlobal: true })
const result = pdm.smartRetrieve('auth patterns', 2000)

ClaudeMdUpdater

import { ClaudeMdUpdater } from '@skillkit/core'

const updater = new ClaudeMdUpdater(projectPath)
updater.update({ minEffectiveness: 70 })

Changes

• Add Claude Code lifecycle hooks (SessionStart, PostToolUse, SessionEnd)
• Implement 3-layer progressive disclosure for token-optimized retrieval
• Add CLAUDE.md auto-updater to populate LEARNED section
• Add auto-compression threshold trigger in observation store
• Cross-platform path handling using path.basename()
• Nullish coalescing for maxResults defaults
• Guard against non-positive tokenBudget
• Thread minRelevance option through findRelevantIds()
• Remove 139 lines of unnecessary inline comments (code cleanup)
• Update documentation with new memory features

Full Changelog: v1.12.0...v1.13.0

v1.12.0 — Runtime Skill Discovery

Runtime Skill Discovery

Adds runtime skill discovery capabilities with REST API, MCP server, pluggable caching, multi-signal relevance ranking, and a Python client.

New Packages

• @skillkit/api — REST API server (Hono) on port 3737

  • GET /search?q=... and POST /search with filters
  • GET /skills/:owner/:repo/:id — get specific skill
  • GET /trending — top skills by relevance score
  • GET /categories — skill categories with counts
  • GET /health and GET /cache/stats
  • Built-in Swagger UI at /docs and OpenAPI 3.1 spec at /openapi.json
  • Sliding-window per-IP rate limiting (60 req/min)

• @skillkit/mcp — MCP server for agent-native discovery

  • 4 tools: search_skills, get_skill, recommend_skills, list_categories
  • 2 resources: skills://trending, skills://categories
  • Stdio transport — works with Claude Desktop, Cursor, any MCP client

• skillkit-client (Python) — async Python client (httpx + pydantic)

  • search(), get_skill(), trending(), categories(), health()

New Core Modules

• Pluggable Cache (@skillkit/core) — MemoryCache with LRU eviction + TTL
• Relevance Ranker — 4-signal scoring (content 40pts, query 30pts, popularity 15pts, references 15pts)
• SKILL.md Parser — frontmatter parsing, reference discovery, content extraction
• Runtime Injector — fetch + cache SKILL.md from GitHub at runtime
• Community Registry — curated registry/SKILLS.md with 26 skills across 12 categories

New CLI Command

• skillkit serve / skillkit server — starts the REST API server
  • --port (-p), --host (-H), --cors, --cache-ttl

Documentation

• Hosted API docs at agenstskills.com/api — interactive Swagger UI
• Rewrote all core docs with clean problem/solution framing and progressive examples
• New pages: REST API, MCP Server, Python Client

Usage

# Start API server
skillkit serve

# MCP server (Claude Desktop / Cursor)
{
  "mcpServers": {
    "skillkit": { "command": "npx", "args": ["@skillkit/mcp"] }
  }
}

# Python client
pip install skillkit-client

Full Changelog

v1.11.0...v1.12.0

v1.11.0

What's New in v1.11.0

Viral Growth Features

Tier 2 — Reduce Friction

• skillkit quick — Zero-friction entry point: auto-detects agents, analyzes project, recommends top 5 skills, one-keypress install
• Badge Generator — Interactive website component to create shareable shields.io skill stack badges for READMEs
• Compatibility Matrix — 16 agents × 12 skill categories with filtering, search, and support scores on the website

Tier 3 — Strategic Positioning

• Repositioned all messaging as "npm for agent skills" (README, package.json, Hero, Features)
• skillkit skillmd validate/init/check — SKILL.md ecosystem commands for the open standard
• Federated Search — skillkit find --federated searches GitHub for external SKILL.md files via code search API
• SKILL.md compatible badge added to README

Security & Reliability Fixes

• isPathInside — Hardened with path.resolve() + path.sep comparison instead of naive string replacement
• skillkit quick install — Sanitizes skill names to prevent path traversal; validates destination stays inside install directory
• skillmd check — Uses path.relative() for cross-platform path display (Windows fix)
• Federated search — AbortController timeout (10s default), RateLimitError on 403 with user-facing guidance
• GitHub API auth — GITHUB_TOKEN/GH_TOKEN sent as Bearer token for higher rate limits (10 → 30 req/min)
• Matrix filtering — Query matching agents preserves all categories (and vice versa)

Stats

• 18 files changed, 1,751 insertions
• All CI checks passing (Node 18/20/22, E2E)

v1.10.0 - Hybrid Search Pipeline

What's New

Hybrid Search Pipeline

A new search system combining vector embeddings with keyword search for improved skill discovery.

Features:

• Vector Embeddings - Semantic search using local GGUF models (nomic-embed-text)
• RRF Fusion - Reciprocal Rank Fusion combines multiple rankers for better results
• Query Expansion - LLM-powered query variations with synonym fallback
• Position-Aware Blending - Different weights for top3, top10, and rest results
• Local Models - Auto-download GGUF models (~2GB), works offline after setup

New CLI Options

# Enable hybrid search
skillkit recommend "authentication" --hybrid

# With query expansion
skillkit recommend "auth" --hybrid --expand

# With LLM re-ranking
skillkit recommend "form validation" --hybrid --rerank

# Build embedding index
skillkit recommend --build-index

New Search Module

File	Description
embeddings.ts	Vector embedding generation
vector-store.ts	SQLite storage with sqlite-vec
rrf.ts	RRF fusion algorithm
expansion.ts	Query expansion with caching
hybrid.ts	Main hybrid search pipeline
local-models.ts	GGUF model management

Technical Details

• 56 new unit tests for search module
• Graceful degradation when embeddings unavailable
• SQL injection protection for table names
• Backpressure handling for large model downloads
• Resource cleanup with try/finally patterns

Full Changelog: v1.9.0...v1.10.0

v1.9.0 - Skill Tree & Reasoning Engine

What's New

Hierarchical Skill Tree

Browse 15,000+ skills organized in a navigable taxonomy with 12 categories.

skillkit tree                    # Full tree view
skillkit tree Frontend           # Category navigation
skillkit tree "Frontend > React" # Subcategory
skillkit tree --stats            # Statistics
skillkit tree --generate         # Generate from index

Categories: Development, Frontend, Backend, Mobile, DevOps, Testing, Security, AI/ML, Database, Tooling, Documentation, Performance

LLM-Based Reasoning Engine

Smart skill discovery using reasoning over the taxonomy:

skillkit recommend --explain     # See WHY skills match
skillkit recommend --reasoning   # Use LLM-based search

Output:

vercel-react-best-practices (Score: 92)
├── Matched: React, TypeScript, Next.js
├── Your stack: Next.js 14, React 18
└── Path: Frontend > React > Best Practices

Connector Placeholders

Tool-agnostic skill authoring with 13 connector categories:

• ~~CRM - Salesforce, HubSpot, Pipedrive
• ~~chat - Slack, Teams, Discord
• ~~email - Gmail, Outlook, SendGrid
• ~~calendar - Google Calendar, Outlook Calendar
• ~~docs - Notion, Confluence, Google Docs
• ~~data - BigQuery, Snowflake, PostgreSQL
• And more...

Execution Flow Tracking

Step-by-step execution with metrics, retries, and progress tracking.

Standalone vs Enhanced Mode

Automatic MCP capability detection with graceful fallback.

Bug Fixes

• Fixed --skill alias and agent detection
• Fixed timer leak in ExecutionManager
• Fixed average stats calculation with cache hits
• Fixed tilde path expansion (use os.homedir())
• Fixed tree traversal for missing nodes

Documentation

• New Skill Tree documentation
• Updated recommendations with reasoning features
• Updated TUI docs with tree view mode
• Updated README and website

Full Changelog

v1.8.1...v1.9.0

v1.8.1 - TUI OpenTUI/Solid.js Migration

SkillKit v1.8.1

🎨 TUI Framework Migration

This release completes a major rewrite of the Terminal UI from React/Ink to OpenTUI/Solid.js, delivering improved performance and a more reactive architecture.

✨ New Features

• OpenTUI/Solid.js Framework: Complete TUI rewrite with reactive primitives
• 20+ New Components: AnimatedText, Button, Breadcrumb, Clickable, CodeBlock, DetailPane, EmptyState, ErrorBoundary, FormField, HoverHighlight, SelectList, SplitPane, StatusIndicator, TabBar, and more
• Service Layer: New modular service architecture for all screens
• Windowed List Navigation: Lists now scroll to keep selection visible
• Enhanced Error Handling: All async operations properly handle errors and show UI feedback

🔧 Improvements

• Security: openUrl now uses execFile instead of exec to prevent command injection
• Robustness: Terminal dimensions clamped to prevent negative values on small terminals
• Render Handling: TUI renderer initialization is now properly awaited with error handling
• Re-entrancy Guards: Prevent concurrent execution of async operations
• Index Clamping: Selection indexes are reactively clamped when lists change

📦 Screens Migrated

All screens have been migrated to the new framework:

• Home, Browse, Installed, Marketplace, Settings
• Recommend, Translate, Context, Memory
• Team, Plugins, Methodology, Plan
• Workflow, Execute, History, Sync
• Help, Mesh, Message

🛠 Technical Details

• Framework: Solid.js with OpenTUI renderer
• Type definitions: New solid-opentui.d.ts with full event handler support
• Build: tsup with ES2022 target
• Runtime: Requires Bun >=1.2.0 for TUI

📚 Links

• Website: https://agenstskills.com
• npm: https://www.npmjs.com/package/skillkit

---

Full Changelog: v1.8.0...v1.8.1