113 skills. Read them all. Install the packs.

Finds and mitigates bot, scraper and datacenter traffic that pollutes analytics or loads an ecommerce site, using Cloudflare's controls.

Sessions spike without revenue, or analytics look wrong on a site behind Cloudflare.

Pulls marketing data from ad platforms, analytics, CRM and commerce APIs, validates it, and reports the source and freshness of every number.

A report, audit or decision needs real account data, not a screenshot or someone's memory.

Plans, audits, instruments and stages Meta campaigns from account evidence, pinned to a named Graph API version, with budgets and protection boundaries set before anything spends.

Launching or restructuring Meta ads and you want the measurement right before the money moves.

Ranks an existing Instagram archive into a reviewed, per-stage shortlist of posts worth paying to distribute, with gates derived from that account rather than borrowed.

You have an organic library that works and you want paid to carry it instead of guessing at new creative.

Plans, audits, duplicates and launches paid Search campaigns from historical evidence, current account configuration, governed budgets and explicit protection boundaries.

Launching or restructuring Google Ads Search without burning budget you can't get back.

Reconciles a founder's bio, positioning and proof across LinkedIn, the website, communities and directories so they all say the same thing.

Your profiles drifted and a prospect could read three different versions of you.

Researches, structures, drafts and reviews website messaging for accounting, tax, legal, financial and other regulated firms, with controls on claims, privacy, testimonials and secure intake.

The firm's website can't say anything it couldn't defend to a regulator.

Runs a research brief from a free prospect deliverable through to a paid strategy and implementation package, with the same structure at every tier.

You need a brief a client can act on, at the depth they've paid for.

Takes an approved strategy through research, visual production, build, GitHub-to-Vercel deployment, QA and handoff, ending in a verified review URL.

A site is approved on paper and needs to become a reviewable, deployed thing.

The full website sequence, end to end: research, strategy, design, build, polish, QA, deploy, present. Each step carries evidence, a credential rule and a gate before the next.

You want a site built the way an agency would build it, with the controls agencies often skip.

Posts, searches, DMs and uploads media on X through its API from the command line.

Publishing or monitoring on X needs to be scripted, not clicked.

Models which model each agent should route to, and what the fleet costs to run at that mix.

You're choosing models per agent, or need a monthly cost estimate before scaling.

Diagnoses and recovers when client-facing responses, scheduled outputs or data sources stop arriving.

Something that used to show up every morning didn't.

Inspects, designs, applies and verifies Cloudflare edge rules without breaking verified bots, checkout flows or analytics.

You need a firewall rule, redirect or cache change on a live store and can't afford to break checkout.

Connects agents to a community platform so they can read, respond and post inside it.

You want an agent present in your community, not just in your inbox.

Keeps tasks consistent when they cross isolated agent runtimes or separate boards.

Two agents on different machines are working the same task list.

Accepts tasks sent into the fleet by an outside authority, records them, and routes them to the right owner.

A client, partner or system needs to hand work to your agents directly.

Works out whether an agent is actually down, blocked or failing auth by separating symptom labels from the real failing layer, using non-destructive live probes.

The board says 'blocked' and nobody knows why.

Stands up and governs a multi-agent fleet to a house standard: one role per agent, the definition contract the loader actually reads, vaulted credentials scoped to a lane, supervised scheduling and approval gates bound to versions.

A second agent is about to join the first and what you need is a standard, not a longer prompt.

Administers user identities and access in SaaS and internal consoles: password resets, invitations, role changes, revocation and account recovery.

Someone joined, left, or locked themselves out.

The format and process for a knowledge base an agent fleet can be trusted to reason from: source registry, extraction, enrichment, framework scoring, and version-bound human review before anything is promoted.

Retrieval keeps returning confident nonsense, or you are about to point an agent at a folder of documents and call it knowledge.

Keeps a conversation that spans several workstreams scoped, so each piece of work stays on its own track.

One thread has turned into four projects.

Inspects, restarts and completes platform migrations on VMs, especially services that were staged but never made operational, with post-migration backfill and health checks.

The migration 'finished' but half the services aren't actually running.

Audits activation checkers, readiness registries, service-manager state, isolation controls and live canaries without changing the target.

Before you trust that a service is up, prove it.

Provisions, diagnoses, restores, verifies and hardens Slack and Telegram coworker agents, including dedicated content agents.

You want an agent that lives in your Slack, or the one you have stopped answering.

Searches arXiv papers by keyword, author, category or ID.

You need the primary source, not a summary of it.

Recovers pages that are blocked, paywalled or behind a firewall through fallback routes.

The page you need to read won't load for an agent.

Monitors blogs and RSS/Atom feeds for new posts.

You want to know the moment a source publishes.

Watches named companies for material news and produces cited digests.

You want a weekly brief on three competitors without reading everything yourself.

Grounds answers and documents in cited, verifiable sources.

The document is going to someone who will check.

Runs Python iteratively in a live Jupyter kernel, keeping state between steps.

An analysis needs to be explored, not scripted once.

Builds and queries an interlinked markdown knowledge base maintained by the model.

Your research is piling up and needs a structure that compounds.

Consolidates financial metrics across businesses and entities into one reconciled view.

You run more than one company and want one P&L conversation.

Inspects a public site's lead gates, forms and funnel routes to map how a visitor becomes a lead.

You want to know how a competitor's funnel, or your own, actually works.

Writes machine-learning papers from design through submission, in conference format.

Rarely. It's here for when the work is publishable.

Triages an inbox: prioritises threads and drafts replies safely.

Morning inbox, forty threads, ten minutes.

Samples an email corpus to calibrate a brand or personal voice without mixing identities or exposing private content.

You want the agent to write like you, learned from how you actually write.

Sends and reads email over IMAP and SMTP from the terminal.

An agent needs a mailbox without a browser.

Designs and maintains agent identity: soul, operating doctrine, fleet roles, memory governance, project instructions, response styles and reusable skills.

You're building an agent that has to behave like a colleague, not a chatbot.

Reads and writes Airtable records through its REST API: filters, upserts, full CRUD.

Airtable is your operating database.

Manages Box cloud files: sharing, search and metadata.

Your documents live in Box.

Extracts cited obligations, deadlines and tasks from a document.

A contract or brief landed and you need what it actually requires.

Creates, reads, edits, templates and reviews Word documents.

The deliverable has to be a .docx.

Works Gmail, Calendar, Drive, Docs and Sheets from the command line or Python.

Automating anything in Google Workspace.

Retrieves and verifies Drive, Docs, Sheets, Calendar and Gmail resources through approved API identities and vaulted credentials.

An agent needs a document, a calendar or a mailbox, under the right identity.

Geocodes, finds points of interest, routes and timezones via OpenStreetMap.

A task needs a location, a distance or a drive time.

Turns meeting notes into cited decisions, owners and tickets.

The meeting ended and nobody wrote down who owns what.

Edits text inside existing PDFs from natural-language instructions.

A PDF needs one line changed and you don't have the source.

Reads and writes Notion pages and databases through its API and CLI.

Notion is where your team works.

Reads, searches, creates and edits notes in an Obsidian vault.

Obsidian is your second brain.

Extracts text from PDFs and scans, including layout-aware conversion.

The information is trapped in a scan.

Creates, reads, merges, fills and secures PDF files.

Any PDF job that isn't editing text in place.

Creates, reads and edits PowerPoint decks.

The deliverable has to be a deck.

Watches product, flight or listing prices and alerts when a target is hit.

You're waiting for a price to move.

Organises agent sessions by prompt: find, rename, archive, prune.

You've run hundreds of sessions and need the one from Tuesday.

Summarises Microsoft Teams meetings, replays jobs and manages Graph subscriptions.

Your meetings happen in Teams.

Runs a weekly reset: commitments, stalled work, next week's plan.

Friday afternoon.

Locates and verifies previously created business documents across repositories, local artifacts and conversation history, without confusing evidence of existence with the actual file.

'I know we wrote this somewhere.'

Creates, reads and edits Excel workbooks and CSVs.

The data has to land in a spreadsheet.

Draws dark-themed SVG architecture, cloud and infrastructure diagrams as HTML.

You need to explain a system to someone who won't read the code.

Renders ASCII art: text banners, boxes, image-to-ASCII.

A terminal deserves a little personality.

Converts video or audio into coloured ASCII MP4 or GIF.

Novelty content, done well.

Designs one-off HTML artifacts: landing pages, decks, prototypes.

You need to see it before you build it.

Generates images, video and audio through diffusion workflows.

Original visuals at volume, on your own hardware.

Authors, validates and exports design-token spec files.

Your brand needs to be machine-readable.

Produces hand-drawn-style Excalidraw diagrams: architecture, flow, sequence.

A whiteboard sketch, without the whiteboard.

Strips AI-isms from text and restores a real voice.

The draft reads like a model wrote it.

Builds infographics from 21 layouts and 21 styles.

A stat needs to travel on social.

Animates math and algorithm explainers in the 3Blue1Brown style.

A concept needs motion to land.

Writes p5.js sketches: generative art, shaders, interactive, 3D.

Something visual and alive for a page or a post.

Reproduces 54 real design systems (Stripe, Linear, Vercel) as HTML and CSS reference.

You want a site that looks like the ones you admire, done properly.

Builds creative browser demos with DOM-free text layout.

Typography as the visual.

Throws together two or three disposable HTML mockup variants to compare.

Deciding a direction before committing to one.

Songwriting craft plus prompts for Suno AI music.

You need a track and don't have a musician.

Controls TouchDesigner through MCP.

Live visuals for an event or installation.

Searches and downloads GIFs from Tenor.

A reaction beats a paragraph.

Downloads and verifies publicly accessible web media safely.

You need the asset, not a link to it.

Produces audio spectrograms and features (mel, chroma, MFCC) from the command line.

Analysing audio, not just playing it.

Turns YouTube transcripts into summaries, threads and blog posts.

An hour of video needs to become ten minutes of reading.

Delegates coding to Claude Code: features, fixes, pull requests.

The build is real and needs a real engineer's workflow.

Delegates coding to OpenAI Codex CLI.

A second coding agent for parallel or comparison work.

Drives the desktop in the background without stealing focus.

The task lives in an app with no API.

Configures, extends and contributes to the Hermes agent runtime.

You're changing how the fleet itself works.

Resolves merge conflicts between agents as a neutral third party.

Two agents edited the same file.

Delegates coding and PR review to OpenCode CLI.

A third coding agent, or a reviewer for the first two.

Runs exploratory QA on a web app: finds bugs, captures evidence, writes the report.

Before a client sees it.

Audits an existing feature end to end: routes, UI, persistence, background writers, authorization, live data shape and tests, without changing code.

'What does this feature actually do today?'

Authors skill files: frontmatter, structure, conventions.

You're writing a skill like the ones on this page.

Writes desktop app plugins that add UI panes and commands.

The agent's desktop needs a new panel.

Reads the live desktop app's DOM and CSS over the DevTools protocol.

Debugging the agent's own interface.

Debugs Node.js through the inspector and DevTools protocol from the CLI.

A Node service is misbehaving and logs aren't enough.

Writes a markdown plan before any execution.

Think first.

Debugs Python with pdb and remote debugpy.

A Python service is misbehaving and logs aren't enough.

Runs a pre-commit review: security scan, quality gates, auto-fix.

Every commit.

Cleans up recent code changes with four parallel agents.

It works, and now it should be readable.

Runs a throwaway experiment to validate an idea before building it.

You're not sure it's possible yet.

Four-phase root-cause debugging: understand the bug before fixing it.

The fix that didn't fix it.

Enforces red-green-refactor: tests before code.

Code you'll have to trust later.

Measures a codebase: lines, languages, ratios.

Sizing a codebase before touching it.

Sets up GitHub auth: tokens, SSH keys, CLI login.

First day on a new machine.

Reviews pull requests with diffs and inline comments.

A PR is waiting on you.

Carries an issue to a verified PR with an honest CI state.

An issue is ready to be built.

Creates, triages, labels and assigns issues.

Managing the backlog.

Runs the PR lifecycle: branch, commit, open, CI, merge.

Shipping a change the right way.

Clones, creates and forks repositories; manages remotes and releases.

Setting up or reorganising repos.

Generates music and sound effects from text with AudioCraft.

You need a sound, not a stock library.

Benchmarks language models on standard evaluations.

Choosing between two models on evidence.

Searches, downloads and uploads models and datasets on Hugging Face.

Getting a model onto your machine.

Runs local GGUF model inference and discovers models on the Hub.

Running a model on your own hardware.

Warehouses large model weights to durable object storage, streamed shard by shard, pinned to a commit and verified, without filling local disk.

You're keeping models you can't afford to re-download.

Segments images zero-shot from points, boxes or masks.

Cutting a subject out of a photo, at scale.

Serves language models at high throughput behind an OpenAI-compatible API.

The fleet needs a local model endpoint.

Logs experiments, sweeps, model registry and dashboards.

Training or fine-tuning anything you'll need to reproduce.