skills

Use when the user wants to drive their real running Chrome with agent-browser-style commands — open pages, snapshot interactive elements as @e1 refs, click, fill, type, screenshot, scroll, eval. Triggers on chrome, browser automation, web automation, fill a form, click a button, snapshot the page, @e1 refs, agent-browser, take a screenshot. Connects to the user's actual Chrome profile (live sessions, cookies, auth) via DevToolsActivePort autoConnect, zero dependencies. Do NOT use for headless/throwaway browsers or launching a new Chrome instance.

Produce a 1-minute visual ops/session report as an HTML artifact: status banner, root-cause table, one flow diagram, shipped/open tables, and a session work log with PR links and live-verified statuses. Use when asked for a report, session report, incident recap page, or 'write up what we did'.

analyse assets using fundamental, technical, smartmoney and narrative. Use when user ask to analyze stocks or an advice if he should buy hold or sell

Use when directing, configuring, recovering, or scheduling a Hermes Agent through chat, Telegram, or another messaging gateway instead of logging into its host. Triggers on "prompt Hermes", "set up Hermes", "configure Hermes cron", "recover Hermes credentials", "message the Hermes bot", "make Hermes do this daily", and "Hermes ignored my prompt".

Write or edit blog posts in the user's preferred style: short, technical, software-engineer voice, understandable to non-experts, value-first, with light SEO and subtle promotion of the relevant product or service. Use when asked to create, draft, rewrite, review, or improve a blog post, article, technical announcement, founder note, or SEO content.

Write or improve any AI agent instruction surface — SKILL.md, AGENTS.md, CLAUDE.md, GOAL.md, dynamic workflow scripts, or any agent documentation. Also refines a user's rough prompt into a production-quality agent prompt. MUST be loaded whenever writing, editing, reviewing, or refining agent instructions. Triggers: "write a skill", "improve SKILL.md", "write agent instructions", "update AGENTS.md", "update CLAUDE.md", "write GOAL.md", "create a workflow script", "write a dynamic workflow", "refine this prompt", "turn this into an agent prompt", "improve this prompt", "prompt engineering". For scaffolding/discovery, pair with skill-creator.

Find a past GitHub Copilot CLI agent session on this machine by project or topic via BM25 search, then summarize what was done. Use when the user asks "which Copilot session did I do X in", "find the Copilot CLI session where I worked on Y", or wants to locate/recall a previous GitHub Copilot conversation. Reads local ~/.copilot session stores. Not for searching code, git history, or the current live session.

Use the preinstalled gws CLI for Google Drive, Gmail, and Calendar workflows, including customer-assisted OAuth login and multi-account control.

Produce a short 3-5 sentence post-mortem when an AI agent's output was wrong or based on faulty verification. Plain English, technical, no headers. Trigger phrases: "retro", "retrospective", "what went wrong", "post-mortem", "reflect on this error", "why did this fail", "the fix didn't work", "what should the agent have done", "analyze this failure".

Operate as an autonomous solo founder / CEO meta-agent driving a whole project — decide what to work on, spawn doer subagents across directions, evaluate and improve them, and integrate, not just finish a known task. Use when handed a project (not a single issue) to push toward a healthy, shipping state; when asked to "run the project", "act as CEO/founder", "make this succeed", "find and fix what matters", "orchestrate agents to deliver this end to end", or to autonomously prioritize and ship across a codebase. For driving ONE known issue to merge, use take-ownership; this skill decides which issue that should be, spawns/evaluates the doers, and hands execution to it.

Take full ownership of a task end-to-end. Use when the user invokes `/take-ownership`, or says "take ownership", "own this", "drive this to merge", "/own", or hands off a task they want completed without micro-management. Walks all phases: issue → design → plan → implement → review → test (no mocks) → PR → CI → final review → merge ask → post-merge prod verify → close.

Manage DNS records for domains registered at Spaceship (spaceship.com). Add/edit/delete A, CNAME, TXT, MX records via the Spaceship Advanced DNS web UI using vibebrowser-cli (VBC). Use when adding subdomains, configuring Vercel custom domains, setting up email forwarding, or any DNS change for agentlabs.cc or other Spaceship-managed domains.

Drain a GitHub pull-request backlog autonomously. Fetches all open PRs, spawns isolated subagents per PR across 5 stages (review, fix, reflect, merge, post-merge CI), logs every stage to backlog.csv. Use when user asks to clear, drain, close, merge, or resolve a PR backlog. Triggers on "clear PRs", "drain pull requests", "merge all PRs", "no PR backlog", "clean up PRs", "resolve open PRs", "close stale PRs". Not for individual PR review (use take-ownership) or issue backlogs (use no-github-backlog).

Send and read Telegram messages and media from a personal Telegram user account using the bundled `telegram-cli.py` Telethon helper. Use for DM-ing a bot or user, reading recent chat history, sending images/files/voice notes, downloading voice messages, and send-and-wait bot smoke tests. Triggers on: "message @SomeBot on Telegram", "read my Telegram chat with X", "send a Telegram message/photo/voice note", "test the Telegram bot reply", "download Telegram voice messages".

Use when starting any task (retrieve relevant context) or finishing any task (persist discoveries). Also use when asked to remember, forget, or look up something from prior sessions.

Create a git worktree for isolated branch work by symlinking node_modules and .env files from the main repo — no reinstall. Use when asked to "create a worktree", "add worktree", "work on a branch without reinstalling", "set up worktree for branch". Never run npm/yarn/pnpm install inside the worktree unless package.json changed on the branch.

Drain a GitHub issue backlog autonomously. Spawns isolated subagents per issue across 7 stages (investigate, implement, review, security-review, qa, fix, merge), logs every stage to backlog.csv. Use when user asks to clear, drain, close, or resolve an issue backlog.

Convene the macro-economist panel — run a market, asset, or portfolio question through multiple thinker-lenses at once and surface their AGREEMENT and DISAGREEMENT instead of one view. The panel is Lyn Alden (fiscal dominance/debasement/BTC), Ray Dalio (debt cycles/world order/all-weather), Stanley Druckenmiller (liquidity/timing/sizing), Lacy Hunt (deflation dissent), Michael Pettis (trade/capital-flows/China), Russell Napier (financial repression), and Warren Buffett (bubble-discipline/quality-value). Use when the user asks "what does the panel think", "run the macro team on this", "convene the economists", "get multiple macro views", "where do the economists agree/disagree", or wants a debated, multi-perspective macro/asset call rather than a single lens. Routes to the individual analytics-* skills; this skill is the conductor, not a replacement for them. Educational, not advice.

Evaluate and improve the `take-ownership` skill from its REAL recent runs. Load this when you want to score how take-ownership performed across recent Claude Code sessions and fine-tune the skill from what actually went wrong. Reads the run registry at ~/.local/run/take-ownership/, evaluates each session, diagnoses the dominant systemic failure, harvests it into a frozen eval case, and runs the DGM-H improve-loop so a meta agent (not you) rewrites SKILL.md. Triggers on "evaluate take-ownership", "improve take-ownership from recent sessions", "how did take-ownership do", "tune take-ownership", "meta-eval take-ownership".

Configure and activate the Claude Max subscription proxy so the agent runs on Claude Opus 4 at flat-rate cost instead of per-token Azure billing. Use when the user says "configure Claude Max", "use my subscription", "set up Claude Max", or asks why the agent is using GPT/Azure instead of Claude.

Measure and improve an existing SKILL.md by running it through a subagent eval loop — actor subagents execute the skill on frozen cases, a judge subagent scores against a rubric, you fix the prompt by general principle and repeat until scores plateau. Use when asked to "evaluate a skill", "improve a skill with evals", "score this skill", "run an eval loop on this skill", "make this skill better", or to tune any prompt/agent instruction against measurable cases. For first-draft authoring use write-skill; this skill improves one that already exists.

Evaluate and improve the `solo-founder` skill from its REAL recent runs. Load this when you want to score how solo-founder performed across recent Claude Code / OpenCode sessions and then fine-tune the skill from what actually went wrong. Reads the run registry at ~/.local/run/solo-founder/, evaluates each session, diagnoses the dominant systemic failure, harvests it into a frozen eval case, and runs the DGM-H improve-loop so a meta agent (not you) rewrites SKILL.md. Triggers on "evaluate solo-founder", "improve solo-founder from recent sessions", "how did solo-founder do", "tune solo-founder", "meta-eval solo-founder".

Find a past Claude Code session on this machine by project or topic, then summarize what was done in it. Use when the user asks "which session did I do X in", "find the session where I worked on Y", "in project ~/path I worked on Z — find and summarize that session", "what was I doing in <project> last week", or wants to locate/recall/analyze a previous Claude Code conversation. Reads local ~/.claude/projects transcripts. Not for searching code, git history, or the current live session.

Use when storing, retrieving, listing, or organizing secrets in Bitwarden using the bw CLI. Triggers on: "store a secret", "get the password for", "save credentials", "list vault items", "create a folder in Bitwarden", "read from Bitwarden", "load API key from vault".

Write, compact, or deduplicate an AGENTS.md / CLAUDE.md project instruction file. Use when creating agent instructions from scratch, or when asked to shorten, dedupe, clean up, or remove fluff/AI-slop from an existing AGENTS.md or CLAUDE.md. Triggers: "write AGENTS.md", "compact our agent instructions", "dedupe CLAUDE.md", "trim the agent rules".

Control OpenCode through the REST API exposed by opencode serve. Use when starting a server, listing sessions, sending prompts, streaming events, handling permission requests, and automating follow-up work from external systems.

Drive a remote chrome-devtools-mcp server (typically on a tailnet) over HTTPS using the chrome-devtools CLI. Use this when the user wants to navigate, screenshot, inspect, or evaluate JavaScript on a browser running on another host (e.g. a Tailscale-connected Mac mini or a CI runner) — and you don't have a local Chrome to control. Examples of triggers ("open <url> on the lab mac", "take a screenshot of the browser on host X", "evaluate this on the remote browser").

Register a free DuckDNS subdomain (*.duckdns.org) and point it at a server IP using the Chrome DevTools browser automation. Use when you need a public DNS name for a dev server and no managed DNS (Cloudflare, Route 53) is available.

Convert Mermaid diagrams (sequenceDiagram, flowchart, graph) into native Excalidraw elements (rectangles, arrows, text) and merge them into an existing `.excalidraw` file. Use when asked to "convert mermaid to excalidraw", "add diagram to excalidraw", "migrate mermaid diagram", "add this mermaid to the mindmap", or "convert .mmd to excalidraw".

Deploy a site with Git LFS assets (videos, images) to Vercel without serving 15-byte pointer files. Use when "videos not loading on Vercel", "LFS files not served", "Vercel serving pointer files", "15 bytes instead of video", "text/plain for video", or "deploy site with LFS assets". Covers the prebuilt-deploy workflow and required GitHub Actions secrets.