Skill Reef 🪸🦞

A growing catalog of reusable OpenClaw skills.

Vision

See VISION.md for the repo direction. Short version: skills should be useful at response time, self-contained enough to copy, and grounded in verified command shapes instead of pointer chains.

Included skills

• asana-formatting - Asana rich-text HTML, task title/body style, comments, mentions, and provenance footers.
• bottom-feeder - Knowledge crawling and note synthesis for durable topic/research files.
• claude-foreman - Dispatch heavy coding work to Claude CLI while OpenClaw orchestrates.
• interactive-sessions - Design and operate button-driven Telegram/Slack chat sessions, games, wizards, polls, and guided workflows.
• knowledge-search - Local semantic knowledge-base search with ChromaDB + Ollama.
• mermaid - Mermaid diagram authoring and rendering workflow.
• moltmaster - Controlled OpenClaw OAuth/auth-profile refresh operations with dry-run, backups, and provider guardrails.
• pdf-to-markdown - Convert PDFs to Markdown with Docling.
• plugin-creator - Create, design, review, and troubleshoot OpenClaw plugins, slash commands, message presentation buttons, and interactive handlers.
• ralph-wiggum - Bounded iteration loop for small projects and prototypes with verification after each slice.
• shell-swap - Mass-switch OpenClaw model settings with codex/gpt/claude lane aliases.
• shop-agent - Browser-driven Amazon/retailer shopping: search, compare, price-check, add to cart, and walk checkout — always stopping for human approval before purchase. Pairs with web-use for product/price data and browser context.
• shrimp - Internal pass-through helper for /shrimp sub-agent dispatch.
• telegram-ui - Telegram rich chat UI patterns: inline buttons, polls, edits, replies, reactions, media, and pins via OpenClaw. Includes live-verified authoring rules for OpenClaw's experimental richMessages mode (see below).
• web-use - Front-door routing for web search, fetch, extraction, protected backends, and real browser context.

⚡ Heads-up: OpenClaw richMessages changes Telegram rendering semantics

OpenClaw ships with Telegram richMessages off by default. Turning it on unlocks tables, collapsible <details> blocks, math, checkbox lists, and inline media — but it also breaks every newline-based formatting habit:

• Plain newlines, blank lines, • bullets, and markdown -/1. lists all collapse into run-on text. The rich renderer treats whitespace like a browser.
• Structure must come from explicit HTML blocks: <p> per paragraph, <ul>/<ol> + <li> for lists, <br> for forced breaks (never <br/> — the self-closing form gets escaped to literal text).
• Empty <p></p> spacers are collapsed too. To get a real blank line between paragraphs, use a paragraph holding a non-breaking space: <p>&#160;</p>.

All of this is live-verified on iOS (2026-07-05) and documented with the full rendering matrix and house authoring rules in skills/telegram-ui/SKILL.md. If your agent's Telegram messages suddenly render as one giant run-on wall after enabling rich mode, this is why.

Companion kits

• x-twitter-kit - X/Twitter access stack for OpenClaw: bundled Peeper no-credit public-account monitoring, xAI/Grok x_search, xurl OAuth setup, diagnostics, and posting safety. Canonical repo: https://github.com/clawSean/openclaw-x-twitter-kit

Companion kits are indexed here for discovery but keep their source in their own repos when they carry larger docs, tests, fixtures, or setup flows.

Notes

• This repo contains the public-safe subset of the local custom skill library.
• Skills with environment-specific secret references or sensitive local bindings are intentionally omitted.

Structure

skills/
  <skill-name>/
    SKILL.md
    references/
    scripts/

Each skill lives in its own directory so it can be copied or installed cleanly into an OpenClaw workspace.