Claude Skills
The open format is called Agent Skills and works in Claude Code, Codex, Cursor and other agents — most people know it as Claude Skills.
Every Agent Skill we could find on GitHub, deduplicated by content. 80 149 files from 1 774 authors, of which 62 489 are unique — the rest is the same skill repackaged into someone else's repository. For each one: what it weighs in tokens, whether it ships runnable scripts, and which MCP servers it needs.
Weekdays, heavy browser lane. Reads the member's own signal sources and saved searches, captures dated buying signals together with the contactable people attached to them, and appends both to the ledgers the outreach queue reads. Read only on every people surface, LinkedIn included. It holds every outbound action unless you released the channel, and it never touches a credential.
Weekly. Audits the paid and measurement setup the member already has by reading it, then assembles the parts that are missing as local files: campaign structure, ad copy, negative keyword seeds, and conversion tracking specifications, each one complete and ready to paste. It creates nothing in an account, saves nothing, activates nothing, and spends nothing, unless you released the channel. This is where the spend stop lives.
Weekdays, file work only, no browser at all. Reads every ledger, queue file, and run record written since its own cursors, turns the member's ticks into sent rows and closed cards, folds the card inbox, re-renders the pipeline board, and writes the short morning brief the member opens first. It names the veto window every day. It holds every outbound action unless you released the channel, and it never touches a credential.
Runs once by hand on the first day and once a month after that, on the first weekday, light browser lane. On the first run it researches the business from its own public surfaces before asking anything, writes the strategy folder, seeds the pipeline, reconciles the schedule table, and registers the recurring jobs. On every monthly run it re-reads the evidence the kit produced, applies what changed to the files it owns, and carries every member written setting across verbatim. It holds every outbound action unless you released the channel, and it never enters a credential.
Weekdays, heavy browser lane. Selects the qualified people who have never been written to, drafts each one a first touch into a dated queue file the member reads, and composes each passing draft into the member's own mailbox as an unsent draft. It verifies by the Drafts count and never by a toast. It holds every outbound action unless you released the channel, and it never touches a credential.
Weekdays, heavy browser lane, two halves in one run. First it reads the member's own mailbox for replies on the threads the ledger says were sent and records them. Only then does it work out who is due a follow up, write each one into a dated queue file, and compose it into the mailbox as an unsent draft. It holds every outbound action unless you released the channel, and it never touches a credential.
Weekly, on a Friday, read only everywhere. Scores the week from the ledgers with a source path beside every number, replays the browser flows the other routines depend on, names one thing to kill and one thing to scale, and files both as cards. It sends only where you released the channel, spends only where you released it, never touches a credential, and never writes a number it did not count out of a named file this run.
Weekdays, heavy browser lane. Reads the member's own saved searches and the named sources for one buyer segment, captures candidate accounts together with the contactable people attached to them, scores every row against the named tests in the qualification file, and writes each one with the evidence that qualified it. Read only on every people surface, LinkedIn included. It holds every outbound action unless you released the channel, and it never touches a credential.
Observe answer visibility for approved buyer questions, audit discoverability and file evidence-backed improvement cards.
Weekly, browser only when a source refuses to be fetched. Counts unpublished calendar entries per property by folding the published ledger against the calendar, and where every property is above its runway threshold it records the counts and exits, which is the normal run. Where a property is low it researches the next block from rank evidence, the topic map, and live result sets, refuses any keyword already targeted, and appends fully specified entries after the last existing block. It is the only writer of the calendar and it never modifies, reorders, or renumbers an existing entry.
Monthly, on the last weekday, light browser lane. Re-tests every named qualification test and every buyer segment against a month of this kit's own ledger evidence, then rewrites both files itself where the evidence disagrees with the assumption they were written on. It records every change in strategy/CHANGELOG.md with the ledger path that justified it. It sends only where you released the channel, never re-queues anyone, never edits the contacted ledger, and never writes a number it did not count out of a named file.
Weekdays, browser only when a source refuses to be fetched. Works the single card the standup marked next: reads the shared publishing standard and the card's specification, pulls the live result set for the primary keyword, reads the pages currently ranking, writes the body, the metadata, the internal links and the outbound authority links, generates and compresses a hero, and leaves a complete draft folder with one ready line. It never opens a publishing surface, never touches a live property, and never touches a credential.
Weekly, browser heavy. Unions every sitemap each property declares without a browser, builds the candidate set from URLs not already in the indexing ledger, then drives the member's own search performance console to inspect each candidate and request indexing inside an account wide allowance. Spends what is left on sitemap health, which is unmetered. It inspects, requests, and views or submits sitemaps inside that console and does nothing else there, ever, and it never removes a sitemap.
Weekly, browser heavy, read only on every screen. Fixes a scoring window that neither double counts nor loses an hour, reads the member's search performance console and analytics property, verifies each range actually applied before reading a figure, reconciles every number against what this Employee published, writes the week's scoreboard and the short rolling state file every other routine reads, and files what it found as cards rather than doing the work itself. It types nothing on any screen, changes no setting, publishes nothing, and never touches a credential.
Weekdays, browser only when the property has no publishing route but a screen. Takes the oldest ready draft, resolves its property, and publishes it by that property's own route: a post file plus a registry entry plus a sitemap update plus a build plus a push, or a publishing surface driven through the flow file it owns. It verifies the live URL by loading it, records the published line, and leaves the draft ready on any failure so tomorrow retries it. It publishes one article to one named property and does nothing else outward, ever.
Monthly, browser conditional. On its first run it discovers the member's properties from the sites they name, writes the three strategy files, creates every ledger and folder the kit reads, files the opening cards, and registers the eight scheduled jobs. Every month after that it re-reads the evidence rather than its own previous conclusions, rebuilds the topic map and the internal link map, corrects any property fact it can prove wrong, and records every change with the path of the evidence that forced it. It holds publishing and spending unless you released the channel, and it never touches a credential.
Weekdays, file work only, no browser at all. Reads every run record since it last ran, folds the published, draft, indexing, and rank ledgers, folds the card inbox into the work board, marks the single card the draft run works today, and writes the short morning brief the member opens first. It holds publishing and spending unless you released the channel, and it never touches a credential.
Weekdays, file work only, no browser at all. Reads every ledger, queue file, and run record, turns yesterday's ticks into facts a machine can count, folds the slot inbox, rewrites the content calendar, and writes the one short brief the member opens first. Its most important line names which slots publish today and how to stop them. It publishes only where you released the channel, never replies, spends only where you released it, and never touches a credential.
Weekdays, heavy browser lane. Opens the member's own signed in profile, post, and notification surfaces for every platform in the channel plan, confirms that everything published since the last run is genuinely live, reads the counts attached to each post, captures every comment, mention, quote, reply, and inbound message into a queue the member answers by hand, and drafts a suggested reply for each one. Read only everywhere and totally read only on LinkedIn. It sends only where you released the channel, never replies, spends only where you released it, and never touches a credential.
Weekdays, browser only where a link genuinely needs verifying. Writes tomorrow's slots, one draft per due slot per destination, from the voice file, the drafting standards, and the dated material captured this morning. It rotates the post framework so no skeleton repeats, enforces the plain text and character budget rules the platforms actually impose, and judges every candidate before it lands. It publishes only where you released the channel, never schedules, never replies, spends only where you released it, and never touches a credential.
Runs once by hand on the first day and once a month after that. On the first run it researches the business from its own public surfaces, reads the member's own already published posts, and builds the voice file from real samples rather than from an idea of a voice, then writes the plan, creates the proof inventory, seeds the opening slots, and registers the recurring jobs. On every monthly run it re reads the evidence the kit produced and rebuilds only what the evidence contradicts. It publishes only where you released the channel, never replies, spends only where you released it, and it never enters a credential.
Weekdays, heavy browser lane. Captures the dated raw material tomorrow's drafts are built from, out of the member's own shipped work, their own site and changelog, their own signed in saved searches, and the places their audience already is. Every line carries a source URL read this run, a date, a verbatim quote, and an expiry. It invents nothing, it is read only everywhere and totally read only on LinkedIn, and it sends only where you released the channel, spends only where you released it, and never touches a credential.
Weekly, on a Friday, read only everywhere. Scores the week from the kit's own ledgers with a source path beside every number, cuts it by platform, framework, pillar, slot time, and post length, rewrites the drafting standards from that evidence and from nothing else, replays the browser flows the other routines depend on, and files exactly one thing to stop and one to do more of. It publishes only where you released the channel, never replies, spends only where you released it, never touches a credential, and never writes a number it did not measure.
Monthly, file work only, no browser at all. Re-reads a month of real change outcomes and rewrites the file that decides what the fix runner may change without asking first. A class the member merged unchanged repeatedly widens by exactly one step. A class they edited, closed unmerged, or that regressed narrows immediately. It never touches the outer boundary, which no volume of evidence can widen.
Weekdays, light browser lane. The one routine in this kit with an outward surface. Takes the slots that are due today, whose drafts were written yesterday and passed the copy check, whose hold box is not ticked, and whose destination the member wrote into the publish allow list themselves, and hands each one to the channel the member configured. It records a receipt the instant each slot lands. It never drives a browser control to publish, never replies, sends only where you released the channel a message, spends only where you released it, and never touches a credential.
Weekdays. Takes the one card the standup marked next and turns it into a reviewable change on a branch, never a deployment. It reads the project's own rules and docs first, makes the smallest change the card describes, runs that project's own gate, writes a change brief with the rollback in one line, and pushes the branch for the member to merge. It never merges, never deploys, never applies a migration, and never rotates a key.
Weekly. Works dependency and supply chain debt one project at a time, sorting every candidate into patch and lockfile only, minor, and anything major or breaking. It bumps the first class on a branch behind that project's own gate and pushes it for the member to merge. It holds the other two with a reason and a card carrying the evidence, so the member decides rather than being handed a red build.
Monthly, and on its first run it is the whole setup, launched by hand. It discovers what the member actually runs by walking their code roots and reading the registrar, the host, and the hosted database, then writes the inventory, the domains file, the starting budgets, and the opening safe fix rules, seeds the board, builds the dashboard, and registers one scheduled job per routine. It creates no account, buys nothing, and provisions nothing.
Weekdays. Checks every project in the inventory from the outside, records one dated observation per project, then reads the production log surfaces since its own cursor and groups every error into a fingerprint so a hundred sightings are one incident. It appends the health ledgers, writes the health digest, and cards the incidents that clear the threshold. It changes nothing, deploys nothing, and never rotates a key.
Weekly. Confirms that the registrar, the host, and the hosted database still match what the inventory says, in one read only pass per project, and names every drift with the exact screen, the exact current value, and the exact intended value. It records that a variable exists and never its value, it fixes nothing inside an account, it buys nothing in any state, and it never rotates a key.
Weekdays, file work only, no browser at all. Reads every ledger and run record since its cursor, resolves which drafted changes actually merged by reading the commit graph rather than a dashboard, closes and reopens incidents on that evidence, folds the card inbox, re-renders the review board, and writes the short morning brief the member opens first. It never merges, never deploys, and never touches a credential.
Weekly. Scores the week from the ledgers with a source path beside every number and never writes a number it did not count in a file this run. It names exactly two things and files both as cards, the one fix worth doing next week and the one thing to stop, and it replays every learned browser flow once, read only, so a drifted selector is found on a Friday rather than on a Tuesday morning a routine needed it.
Hire one of the eight open source AI Employees (GTM Engineer, SEO, Web Dev, Social Media, Ad Manager, Sales, Customer Satisfaction, Chief of Staff). Copies that one employee folder from the ai-employees repo to a folder the user chooses outside cloud sync, runs its three self tests, and prints its install prompt with the path filled in. Use when the user says hire an employee, install an AI employee, set up the GTM Engineer, or names one of the eight. It registers no schedule and sends nothing.
> session. Trigger on any `git commit` activity visible in the session — either commits you make yourself, or commits the user makes outside your tool calls that surface in the conversation — even without an explicit request. Also trigger when the user explicitly asks to tag existing commits in a branch (e.g. "add assisted-by to commits in this branch").
> Flux CD and Flux Operator expert — answers questions and generates schema-validated YAML for all Flux CRDs (not repo auditing or live cluster debugging). Use when users ask about Flux concepts, want manifests for HelmRelease, Kustomization, GitRepository, OCIRepository, ResourceSet, FluxInstance, or any Flux resource. When user needs guidance on GitOps repository structure, bootstrap Flux with Terraform, multi-tenancy, OCI-based delivery, image tag automation, drift detection, preview environments, notifications, or the Flux Web UI and MCP Server.
> Run the upstream Flux controller patch release procedure for helm-controller, image-automation-controller, image-reflector-controller, kustomize-controller, notification-controller, source-controller, and source-watcher. Use when preparing a new controller patch release from a release series branch, drafting changelog entries, tagging releases, and opening the follow-up changelog PRs back to main.
> Audit and validate Flux CD GitOps repositories by scanning local repo files (not live clusters) — runs Kubernetes schema validation, detects deprecated Flux APIs, reviews RBAC/multi-tenancy/secrets management, and produces a prioritized GitOps report. Use when users ask to audit, analyze, validate, review, or security-check a GitOps repo.
> Run the upstream Flux controller minor release procedure for helm-controller, image-automation-controller, image-reflector-controller, kustomize-controller, notification-controller, source-controller, and source-watcher. Use when branch, drafting the minor changelog, tagging, merging the release branch back to main, and adding the backport label.
> Debug and troubleshoot Flux CD on live Kubernetes clusters (not local repo files) via the Flux MCP server — inspects Flux resource status, reads controller logs, traces dependency chains, and performs installation health checks. Use when users report failing, stuck, or not-ready Flux resources on a cluster, reconciliation errors, controller issues, artifact pull failures, image automation not updating tags, alerts or webhooks not being delivered, or need live cluster Flux Operator troubleshooting.
100+ animated React components (Aceternity UI) for Next.js with Tailwind. Use for hero sections, parallax, 3D effects, or encountering animation, shadcn CLI integration errors.
REST API security hardening with authentication, rate limiting, input validation, security headers. Use for production APIs, security audits, defense-in-depth, or encountering vulnerabilities, injection attacks, CORS issues.
MUI Base UI unstyled React components with Floating UI. Use for accessible components, Radix UI migration, render props API, or encountering positioning, popup, v1.0 beta issues.
AutoAnimate (@formkit/auto-animate) zero-config animations for React. Use for list transitions, accordions, toasts, or encountering SSR errors, animation libraries complexity.
Use for Docker with Bun, Dockerfiles, oven/bun image, containerization, and deployments.
This skill should be used when the user asks about "Cloudflare Workers with Bun", "deploying Bun to Workers", "wrangler with Bun", "edge deployment", "Bun to Cloudflare", or building and deploying applications to Cloudflare Workers using Bun.
Skill for integrating Better Auth - comprehensive TypeScript authentication framework for Cloudflare D1, Next.js, Nuxt, and 15+ frameworks. Use when adding auth, encountering D1 adapter errors, or implementing OAuth/2FA/RBAC features.
This skill should be used when the user asks about \"bun:ffi\", \"foreign function interface\", \"calling C from Bun\", \"native libraries\", \"dlopen\", \"shared libraries\", \"calling native code\", or integrating C/C++ libraries with Bun.
Use for Bun file I/O: Bun.file, Bun.write, streams, directories, glob patterns, metadata.
This skill should be used when the user asks about "Next.js with Bun", "Bun and Next", "running Next.js on Bun", "Next.js development with Bun", "create-next-app with Bun", or building Next.js applications using Bun as the runtime.
Use when running Nuxt 3 with Bun runtime, building Vue/Nuxt apps with Bun, or configuring Nuxt projects to use Bun for development and production.
Bun package manager commands (install, add, remove, update), workspaces, lockfiles, npm/yarn/pnpm migration. Use for dependency management with Bun.
Use for bun:sqlite, SQLite operations, prepared statements, transactions, and queries.
Use when building or running SvelteKit apps on Bun, including SSR, adapters, and Bun-specific APIs
Use for bun:test syntax, assertions, describe/it, test.skip/only/each, and basic patterns.
TanStack Start full-stack React framework with Bun runtime. Use for TanStack Router, server functions, vinxi, or encountering SSR, build, preset errors.
Use for test lifecycle hooks: beforeAll, afterAll, beforeEach, afterEach, fixtures, preload.
Cloudflare Cron Triggers for scheduled Workers execution. Use for periodic tasks, scheduled jobs, or encountering handler not found, invalid cron expression, timezone errors.
Cloudflare Browser Rendering with Puppeteer/Playwright. Use for screenshots, PDFs, web scraping, or encountering rendering errors, timeout issues, memory exceeded.
Cloudflare Email Routing for receiving/sending emails via Workers. Use for email workers, forwarding, allowlists, or encountering Email Trigger errors, worker call failures, SPF issues.
Cloudflare D1 serverless SQLite on edge. Use for databases, migrations, bindings, or encountering D1_ERROR, statement too long, too many requests queued errors.
Claude Skills — questions
Answers built from the skills we actually parsed.