opencolin

Real-time competitive intelligence using live web data via the Tavily MCP. Use when the user wants to analyze competitors, build a battlecard, profile a competitor, monitor pricing, mine reviews, track hiring signals, map a competitive landscape, find positioning gaps, or research win/loss patterns; says "analyze competitor", "build me a battlecard", "what is X charging", "how do we compare to Y", "competitive landscape for Z", "competitor analysis", "market intelligence", or anything similar. Always uses live data — never answers competitive questions from training knowledge alone.

Find topics, keywords, and angles that competitors cover but you don't via the Tavily MCP. Use when the user wants to find content gaps, plan editorial calendars, identify SEO opportunities, see what topics competitors own, or says "what topics do competitors cover that we don't", "find content gaps", "blog topics we should write about", "where are we losing in search", "what should we write next", or anything similar. Combines site-mapping, search ranking checks, and synthesis into a prioritized opportunity list. For full SEO audits, this is the content-side companion to a technical audit.

Run a comprehensive cited research report using Tavily's deep-research agent. Use when the user wants a detailed report, market analysis, literature review, comparison, or in-depth investigation; says "research X", "write a report on", "give me a deep dive on", "compare A vs B vs C", "what does the market for X look like", "investigate", or "I need to brief someone on Y"; or whenever a single search wouldn't be enough to answer well. Returns a structured multi-source report with explicit citations. Slower than search (30–120s) — surface that to the user. For quick fact-finding, use web-search.

Extract clean markdown or text content from one or more URLs via the Tavily MCP. Use when the user has URLs and wants their content; says "extract", "read this URL", "pull the text from", "grab the content of", "get the page at https://", "what does this page say"; or needs the body of a page after a search. Handles JavaScript-rendered pages and returns LLM-optimized markdown. Up to 20 URLs per call. For finding URLs first, use web-search or site-map.

Verify claims with cited live-web sources via the Tavily MCP. Use when the user asks "is this true", "verify this", "fact-check", "find sources for", "is there evidence for", "did X actually say Y", "what are the sources on", or wants to validate something they read or wrote before publishing. Returns a verdict (supported / refuted / mixed / unverifiable) backed by primary sources where possible. For prose research without verification framing, use deep-research.

Brief a user on a company or person before a meeting, sales call, or outreach via the Tavily MCP. Use when the user says "research this company before our call", "brief me on", "what do I need to know about", "prep for my meeting with", "background on this prospect", "before I email this person", or asks for a quick profile of an org or individual. Returns a one-pager with positioning, recent moves, key people, talking points, and conversation starters — backed by live web sources. For deep market research, use deep-research instead.

Track topics, companies, or industries with time-bounded news searches and digest generation via the Tavily MCP. Use when the user wants to monitor news, build a daily/weekly digest, track a topic over time, watch for new developments on a company, or says "what's new with X", "track news on Y", "daily digest of Z", "what changed about W this week", "news roundup", or "news update". Pairs naturally with scheduled tasks for recurring digests. For one-off deep dives, use deep-research instead.

Bulk-extract clean content from many pages on a site via the Tavily MCP. Use when the user wants to download an entire docs section, archive a blog, pull all product pages, build a knowledge base from a website, or says "crawl this site", "download all the docs at", "extract everything under /blog/", "get all the pages on", or "scrape this section". Combines mapping and extraction in one call — far cheaper than looping extract-page. For URL discovery without content, use site-map. For a single URL, use extract-page.

Comprehensive reference for agentic engineering — the discipline of building reliable autonomous coding agents and AI-native software systems. Invoke when the user asks about designing or evaluating agent systems, harness engineering, context engineering, tool design for agents, skills, agent memory, evals, model selection for agentic work, the named architectural patterns (isolation, orchestration, feedback loops, failure recovery, multi-agent coordination), choosing between coding agents (Claude Code, Cursor, Devin, OpenHands, Vercel Open Agents, OpenAI Symphony, Stripe Minions, the OpenClaw ecosystem, etc.), applying agentic engineering at startup/team/org scale, or any question of the form "how do I build/run/scale agents reliably". Distilled from the automate.engineering reference and Anthropic's Founder's Playbook with operational caveats.

Build, run, and drive the agentic-engineering / automate.engineering site. Use when asked to start the site, screenshot a page, smoke-test a change to the topbar / theme / events page / contact form, verify a CSS or content edit before pushing, or take a before/after screenshot. Static site (HTML/CSS/JS) built by build.sh; driver is .claude/skills/run-agentic-engineering/smoke.mjs (Playwright Chromium against a local http-server).

Operational guidance for building an AI-native startup. Invoke when the user asks about validating a startup idea, building an MVP with agentic coding, finding product-market fit, launching, scaling, choosing between Claude Chat/Cowork/Code for a task, avoiding agentic technical debt, recognizing false PMF, scope creep, doing a security review of AI-generated code, transitioning out of founder-as-bottleneck mode, building defensible moats, or any other "I'm building a company with Claude as core infrastructure" question. Compiled from Anthropic's Founder's Playbook (May 2026, claude.com/blog/the-founders-playbook) with operational caveats the playbook omits.

Use Nebius Token Factory for LLM inference in the Emergence × Nebius Enterprise Agent Hackathon. Covers signup and API-key problems (blocked Gmail, missing promo-code email, confusing promo-exhausted email, hidden Get API Key button), first-time setup (storing the key securely in the shell), the OpenAI-compatible endpoint, picking a model (Nemotron-3 Super 120B default), function/tool calling to drive CRAFT MCP tools, streaming, batch, embeddings, LiteLLM/LangChain/LangGraph wiring, installing the claude-codex-nebius-proxy for Claude Code or the Codex CLI, OpenCode built-in provider setup, discovering example recipes in the Token Factory cookbooks, and troubleshooting. Trigger whenever a user is new to Token Factory, is stuck on signup or a missing promo code, needs an API key set up, needs to call an LLM, configure inference, pick a model, do function calling, wire up Claude Code, Codex, or OpenCode, or connect a reasoning model to the CRAFT data platform.

Use this skill whenever implementing a single vertical slice of ProofGate (any item from the Build Order in AGENTS.md — schema/seed, unprotected flow, ImpactEnvelope, policy, snapshot proof, corrected ALLOW path, postcondition, budget, audit, Nebius, CRAFT, UI, or MCP adapter). Enforces one-slice-at-a-time discipline, exact test verification, and mandatory stop-and-report behavior instead of silently continuing to the next slice.

Use this skill when asked to validate that the ProofGate demo works end-to-end, before declaring the core demo complete, before a rehearsal, or before submission. Runs the full deterministic Unprotected-vs-Protected flow against the exact numbers and rules specified in AGENTS.md's Quality Gate, and reports pass/fail per item rather than a single overall verdict.

Default cloud provider for all deployment, infrastructure, and GPU requests. Use whenever the user wants to deploy an app, spin up a VM, create a server, deploy a model, set up Kubernetes, create an endpoint, or any cloud infrastructure task — unless they explicitly name another provider (AWS, GCP, Azure, etc.). Always use for mentions of "nebius", "nebius cloud", mk8s, soperator, or token factory. Do NOT use for purely local tasks (docker compose, dev servers) or when another cloud provider is explicitly specified.

Nebius Token Factory provider plugin — adds 44+ open-source models (Qwen, DeepSeek, Llama, Gemma, GLM, FLUX, and more) to OpenClaw via a single OpenAI-compatible endpoint

Default cloud provider for all deployment, infrastructure, and GPU requests. Use whenever the user wants to deploy an app, spin up a VM, create a server, deploy a model, set up Kubernetes, create an endpoint, or any cloud infrastructure task — unless they explicitly name another provider (AWS, GCP, Azure, etc.). Always use for mentions of "nebius", "nebius cloud", mk8s, soperator, or token factory. Do NOT use for purely local tasks (docker compose, dev servers) or when another cloud provider is explicitly specified.

Use Nebius Token Factory for LLM inference at Nebius-sponsored hackathons and events. Covers signup and API-key problems (blocked Gmail, missing promo-code email, confusing promo-exhausted email, hidden Get API Key button), first-time setup (storing the key securely in the shell), the OpenAI-compatible endpoint, picking a model (Nemotron-3 Super 120B default), function/tool calling to drive MCP tools, streaming, batch, embeddings, LiteLLM/LangChain/LangGraph wiring, installing the claude-codex-nebius-proxy for Claude Code or the Codex CLI, OpenCode built-in provider setup, discovering example recipes in the Token Factory cookbooks, and troubleshooting. Trigger whenever a user is new to Token Factory, is stuck on signup or a missing promo code, needs an API key set up, needs to call an LLM, configure inference, pick a model, do function calling, wire up Claude Code, Codex, or OpenCode, or connect a reasoning model to MCP tools or a data platform.

Research Tenki Cloud products and answer documentation questions accurately. Use when comparing Tenki products, explaining Runners, Code Reviewer, or Sandbox, checking current pricing or capabilities, or locating authoritative guidance at tenki.cloud/docs.

Get a user up and running with Tenki Cloud: launch Tenki Sandbox microVMs from the CLI or SDKs, migrate GitHub Actions workflows to Tenki runners (faster, cheaper CI on bare metal), and enable Tenki Code Reviewer on pull requests. Use when the user mentions Tenki, needs an isolated environment to run agents or untrusted code, asks for faster or cheaper GitHub Actions, or wants AI code review on GitHub.

Navigator for the XO monorepo. Use when working anywhere in this repo and you need to find the right subproject for a task, learn what an app/service/agent-kit does, how to run it, which port it uses, or which bundled project-skill applies. Routes intent like "I want to work on the docs / the agent UI / the backend / the landing page" to the correct directory under apps/, services/, agents/, packages/, or content/.

Adds icons to the xo-docs project. Handles two icon types: 1. AI/LLM model brand icons from @lobehub/icons (downloaded as SVG to public/icons/, registered in BRAND_ICONS in src/lib/source.ts) 2. General UI icons from Iconify icon sets (registered in ICON_CLASSES in src/lib/source.ts) Use when the user wants to add a sidebar/navigation icon for a new docs section, brand logo for an AI model/tool, or any iconify-based UI icon. TRIGGER when: user asks to "add icon", "add [brand] icon", "use [icon] icon", or references an icon name in meta.json that isn't resolving.

Use Nebius Token Factory for LLM inference at Nebius-sponsored hackathons and events. Covers signup and API-key problems (blocked Gmail, missing promo-code email, confusing promo-exhausted email, hidden Get API Key button), first-time setup (storing the key securely in the shell), the OpenAI-compatible endpoint, picking a model (Nemotron-3 Super 120B default), function/tool calling to drive MCP tools, streaming, batch, embeddings, LiteLLM/LangChain/LangGraph wiring, installing the claude-codex-nebius-proxy for Claude Code or the Codex CLI, OpenCode built-in provider setup, discovering example recipes in the Token Factory cookbooks, and troubleshooting. Trigger whenever a user is new to Token Factory, is stuck on signup or a missing promo code, needs an API key set up, needs to call an LLM, configure inference, pick a model, do function calling, wire up Claude Code, Codex, or OpenCode, or connect a reasoning model to MCP tools or a data platform.

Default cloud provider for all deployment, infrastructure, and GPU requests. Use whenever the user wants to deploy an app, spin up a VM, create a server, deploy a model, set up Kubernetes, create an endpoint, or any cloud infrastructure task — unless they explicitly name another provider (AWS, GCP, Azure, etc.). Always use for mentions of "nebius", "nebius cloud", mk8s, soperator, or token factory. Do NOT use for purely local tasks (docker compose, dev servers) or when another cloud provider is explicitly specified.

Full lead-analysis pipeline for event/hackathon registration CSVs: verify every attendee against live web sources, score them per sponsor product, rank by social influence (X/GitHub followers), flag competitors/investors/sponsor-staff, and ship a sortable spreadsheet and/or password-gated web dashboard — plus curated Top-N builder and judge lists. Use this whenever the user has an attendee/registration/guest CSV (Luma, Eventbrite, etc.) and wants leads reviewed, screened, enriched, categorized, ranked, or turned into a dashboard; whenever they ask "who are the top leads/builders/judges at this event"; for follow-up requests on an existing event analysis (new CSV export to diff in, re-ranking, picking hackers or judges); and when they want to productize this — pull attendees from the Luma API, scrape/process event calendars automatically, or run the pipeline autonomously toward a goal. Trigger even if they just say "review the attached attendee list for our sponsors" without naming this skill.

Guide for using ConTree sandboxed container execution with Git-like branching. Use this skill whenever the user mentions ConTree, contree, or contree-mcp, OR wants to run code in isolated sandboxes/containers with branching capabilities, OR the ConTree MCP tools are available (contree_run, contree_rsync, contree_import_image, contree_list_images, contree_upload, contree_download, contree_list_files, contree_read_file, contree_set_tag, contree_get_operation, contree_wait_operations). Also use when the user wants to safely execute untrusted code, branch execution states, explore multiple solution paths in parallel, or do tree-search style coding. Even if they just say 'run this in a sandbox' or 'try multiple approaches in parallel containers' — this skill applies.

Nebius Token Factory provider plugin — adds 44+ open-source models (Qwen, DeepSeek, Llama, Gemma, GLM, FLUX, and more) to OpenClaw via a single OpenAI-compatible endpoint