← ClaudeAtlas

aiaddonlisted

Use whenever someone wants to build, update, audit, version, or package a Claude plugin, skill, connector, agent, hook, MCP server, or add-on, including internal tools for a team or business. Triggers: "build me a plugin", "make me a skill", "add a connector", "create an agent", "turn this workflow into a Claude plugin or skill", "automate our process for my team", "make Claude check every document against our rules", "update my plugin", "audit my extension", "audit this plugin folder", "what's wrong with this plugin or skill", "bump the version", "package this for Cowork", "write me a prompt", "tighten this prompt", "make this prompt reusable with variables", "clean up my plugin", "find duplicate content in my skills", and mentions of /skill-creator, /create-cowork-plugin, /cowork-plugin-customizer. Entry point and router for AI Add-on Builder. Load first; routes only.
syncwin/aiaddon · ★ 0 · AI & Automation · score 75
Install: claude install-skill syncwin/aiaddon
# AI Add-on Builder Router only. Standards for every extension built or changed: `references/standards.md`. Platform layering: `references/platforms.md`; agent logic shared across platforms: `references/agent-protocols.md`; the Prompt Builder companion prompt every extension ships: `references/companion-prompt.md`. ## Hard rules (every stage, no exceptions) - **PUBLIC SAFETY GATE:** nothing this plugin builds, edits, packages, pushes or publishes may contain secrets, keys, tokens, passwords, personal emails, phones or addresses, client names, internal URLs, local user paths or other private data. Run `aiaddon-review/references/scan-secrets.md` and `public-safety-scan.md` on the folder (and on git history before any push or release) and stop on any finding. Never bypass it. - **COMPLIANCE:** if a request would break Claude or Anthropic official rules (usage policy, plugin directory policy, plugin docs), say so plainly and offer the safe official method instead of building a non-compliant product. Example: the directory refuses hooks, scripts and other non-text files from non-Verified organizations, so offer skill instructions, `userConfig` with `sensitive: true` or a remote MCP over HTTP first. ## 0. Start simple The user should never need to know what a manifest, frontmatter, or MCP is. Talk about what the result will do, not its files. - **Invoked with no request, or the request is vague:** ask one short question and stop: > What would you like Claude to do for you?