lzlrd/marketplace

A four-role dev crew (Planner, parallel Coders and Testers, and a fan-out Reviewer) that builds and reviews features — or stands up a whole new codebase — in the requestor's own style, run as a Claude Code agent team. Auto-engages for feature and new-project work; also exposes /4man:code-review. Requires the experimental agent-teams feature (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS).

Deduce which locally-installed Claude Code skills can run in Claude Desktop / claude.ai and package the qualifiers into upload-ready zips. Inventories plugin and personal skills correctly (latest version, symlinks resolved, no marketplace dupes), classifies each against Desktop's sandbox and MCP list, and curates clean zips with SKILL.md at every folder root. Lenient by default; pass --strict for a zero-code-execution reading. Auto-engages when you want to move skills to Claude Desktop, or run it as /export-skills.

Plan with Fable 5, execute with Opus: an opusplan-style dual-model setup for Claude Code. Tracked as a submodule of the lzlrd/fableplan fork.

Fetches current, version-accurate documentation for any third-party library, framework, SDK, API, or CLI tool via Context7, with a web fallback when it isn't connected. Auto-engages whenever a question is anchored to a specific library or framework.

Wires the mempalace memory server into Claude Code: three hooks that load palace context, pull the user's writing/coding voice, and nudge the session diary, plus a /match-style skill and a /remember handoff for cross-session continuity. MCP-first, with a mempalace-CLI fallback when the MCP isn't connected, and no auto-mining.

Bundles the offshore MCP server. Forwards refused prompts to a local abliterated model (Ollama, LM Studio, vLLM, or any OpenAI-compatible endpoint) and returns the unrestricted reply.

Grounds platform-API answers in live official documentation — Apple docs, Microsoft Learn, and Google's developer knowledge base — when building for macOS, iOS, iPadOS, watchOS, tvOS, visionOS, Windows, or Android. Auto-engages for native platform-SDK and OS-API work.

Turn a description into an optimized, ready-to-paste LLM prompt for a Gemini Gem, Claude Project, or custom GPT, grounded in the promptingguide.ai guide — a bundled distilled reference for offline use, plus the NotebookLM MCP for live sources when connected. Auto-engages when you write or refine a prompt.

Bundles the realfavicon-mcp MCP server: generate favicon assets and HTML from a master image, audit a live page's favicon setup, and query the RealFaviconGenerator change-log.

A /reload-claude-md command that re-reads the in-effect CLAUDE.md memory files from disk and re-anchors the session to their current content, so mid-session edits take effect without a restart. Claude Code loads CLAUDE.md only at session start and has no native reload; this fills that gap. Read-only and user-invoked.

Semantic Versioning 2.0.0 decision procedure: classifies a change as MAJOR, MINOR, or PATCH and produces the exact next version number, grounded in a bundled distilled copy of the spec. Auto-engages whenever a version is being bumped or a release is being cut.

Cloud-agnostic system design across AWS, Azure, Google Cloud, and Cloudflare: a six-step Well-Architected loop, grounded in each cloud's live documentation MCP when connected. Auto-engages for architecture and service-selection work.