skills-by-yigitkonur

skills for ai coding agents — one pack, 54 skills + the internet-researcher agents. review, research, writing, ui/ux audit, mcp & framework builders, frontend/backend testing, browser/device/terminal automation, config files, publish. install what you need, skip the rest. no monolith.

used to be two repos (a main pack + a -secondary b-side). they're one now. the old secondary repo is gone — everything lives here.

install

three ways in. Codex and Claude Code both get the complete 54-skill pack.

as claude code plugins (the good way — toggle on/off via /plugin)

add the marketplace once:

/plugin marketplace add yigitkonur/skills-by-yigitkonur

then grab exactly what you want — one skill, a themed bundle, the researcher agents, or everything — and drop it just as fast:

/plugin install run-review@yigitkonur          # one skill
/plugin install yk-mcp@yigitkonur              # a themed bundle
/plugin install yk-researchers@yigitkonur      # just the internet-researcher agents
/plugin install yk-everything@yigitkonur       # the whole thing
/plugin uninstall run-review@yigitkonur        # gone

installed ≠ enabled. /plugin lets you flip stuff on and off without reinstalling — enable what you're working with, mute the rest so it doesn't eat your context. per-skill plugins are named after the skill (<skill>@yigitkonur); bundles are the yk-* names below.

as a codex plugin

add the marketplace once:

codex plugin marketplace add yigitkonur/skills-by-yigitkonur

then open /plugins in Codex and install the skill you need, such as run-review@yigitkonur; skills-by-yigitkonur@yigitkonur installs the full pack.

Codex offers the same per-skill choice: install run-review@yigitkonur, build-mcp-server-sdk-v2@yigitkonur, or any other skill directly from /plugins. skills-by-yigitkonur@yigitkonur remains the all-pack option for existing users. Codex packages are generated under plugins/, so each install is self-contained; Claude Code also offers its themed yk-* bundles.

with the skills cli

project install (recommended & PromptScript-compatible — installs locally to ./.agents/skills):

npx -y skills add yigitkonur/skills-by-yigitkonur -y                      # full pack
npx -y skills add yigitkonur/skills-by-yigitkonur/skills/<skill-name> -y  # single skill

PromptScript projects can also import skills directly via the prs CLI:

prs skills add github.com/yigitkonur/skills-by-yigitkonur/skills/<skill-name>/SKILL.md

global install (user-level — installs cleanly to ~/.agents/skills across universal agents):

npx -y skills add yigitkonur/skills-by-yigitkonur -y -g -a universal                      # full pack
npx -y skills add yigitkonur/skills-by-yigitkonur/skills/<skill-name> -y -g -a universal  # single skill

per-skill one-liners live in each Claude-compatible skill's README.md; Codex-only skills document the Codex all-pack install instead.

staying fresh

every push to main auto-bumps the version, so /plugin marketplace update always pulls the latest skills and agents. you don't chase releases — they chase you.

bundles

Claude Code themed groups for one-shot installs. every Claude-compatible skill also installs on its own; Codex-only skills are listed separately below.

🏗️ build apps & frameworks

write app code against a specific framework or sdk.

• build-chrome-extension — chrome mv3: manifest v3, service_worker, content_scripts, popup, side_panel, declarativenetrequest, web store packaging.
• build-cloudflare-access-sso — cloudflare access + google sso on a subdomain: reusable policies, google oauth client/consent-screen traps, proxied dns, origin ip lockdown + jwt validation so access can't be bypassed, four-check verification.
• build-cloudflare-email-sending — cloudflare email service: send_email worker binding, wrangler domain onboarding, migrate from resend/ses/postmark, replace supabase auth's mailer.
• build-effect-ts-v3 — effect-ts v3: Effect.gen, Layer, Schema, typed errors, fibers, Stream.
• build-kernel-ts-sdk — kernel sdk (@onkernel/sdk): browsers, apps, profiles, managed auth, pools, playwright/cdp.
• build-langchain-ts-app — langchain.js: agents, tool-calling, rag retrievers, structured output, streaming, langgraph.
• build-licenseseat-swift — licenseseat swift sdk on macos: source-verified api surface (readme snippets that don't compile), two-singleton trap, .pending activation trap, offline grace, seats, hardened licensemanager blueprint.
• build-raycast-script-command — raycast script commands (.sh/.py with @raycast.* header): fields, modes, arguments, discovery.
• build-sentry-macos-swift — sentry-cocoa on macOS/swift: explore-repo → support matrix → deep integration (crash, nsexception, breadcrumbs, tags, scope, tracing, release health, dSYM, privacy scrubbing).
• build-tinacms-nextjs — tinacms + next.js app router: tina/config.ts, mdx/git content, schema modeling, useTina visual editing.

/plugin install yk-build@yigitkonur

🔌 mcp & agent interfaces

build, test, convert, and audit mcp servers, clients, and agent-facing clis.

• build-mcp-server-sdk-v1 — mcp server on @modelcontextprotocol/sdk v1.x: single-package, zod, McpServer.
• build-mcp-server-sdk-v2 — mcp server on @modelcontextprotocol/server v2 alpha: split packages, registerTool, ctx.mcpReq.
• build-mcp-use-server — mcp-use v2: MCPServer tools, views (MCP Apps), oauth providers, streamable HTTP, deploys, v1 migration.
• build-mcp-use-client — mcp-use client: MCPClient, MCPSession, useMcp, mcp-use/react, browser transport.
• build-mcp-use-agent — mcp-use MCPAgent: an llm picks & orchestrates mcp tools via run, stream, streamEvents.
• build-clean-mcp-architecture — clean architecture layer boundaries for typescript mcp-use/server code, import direction, dependency-cruiser gates.
• convert-mcp-sdk-v1-to-v2 — port a v1 mcp server to the v2 split-package sdk: package renames, ServerContext, zod v4.
• test-by-mcpc-cli — drive the mcpc cli (0.6.x) to test/debug/smoke-check an mcp server over stdio or streamable http.
• audit-agentic-mcp — audit an mcp server for agent-readiness, or design a new one before code: framework, security, context posture.
• audit-agentic-cli — audit/design a cli for agent consumption: stable json, exit codes, non-interactive flags, repair loops.

/plugin install yk-mcp@yigitkonur

🧪 frontend & backend testing

author, run, diagnose, and release-gate deployed browser and API tests.

• run-testsprite-backend — TestSprite backend API tests with secure, revision-pinned release proof.
• run-testsprite-frontend — TestSprite browser tests via public CLI or localhost MCP.

/plugin install yk-testing@yigitkonur

⚙️ config & instruction files

generate or refresh the config / instruction files another tool reads.

• init-agent-config — agents.md / claude.md / review.md hierarchies; folder-scoped guidance; native review adapters.
• init-jean-json — onboard a repo to jean: author jean.json + .worktreeinclude, prove them in a throwaway worktree, document in agents.md, retire test worktrees.
• init-makefiles — scaffold safe scenario makefiles (local dev, tunnels, deploys, r2 sync, supabase, railway, vercel, mac shipping).
• update-agent-config — audit agents.md / claude.md / review.md for drift after refactors; refresh refs, recount tables, map folder coverage, fill only invariant-dense gap folders.

/plugin install yk-config@yigitkonur

🎨 frontend rebuild & audit

rebuild a live site, rip its design, or audit a running ui/ux.

• convert-url-to-nextjs — rebuild a deployed site as-is pixel-faithful as a next.js project from a live url — the "we lost the frontend repo" recovery. l0+l1 crawl + back-to-back agent-browser verification.
• optimize-nextjs-fluidity — audit + optimize a next.js app router repo for performance and fluidity, gating every practice against the installed next version, planning one task per file in nextjs-enhancement/, then executing the reversible ones.
• audit-ux-laws — audit ui against the 30 laws of ux (fitts's, hick's, miller's, jakob's, gestalt, choice overload, cognitive load) with critical/minor severity + code fixes.
• audit-ui-and-save-files — visual ui audit across pages/viewports with browser screenshots, per-bug findings to css-issues/[yy-mm-dd]/..., ending with an approval-gated fix-subagent plan.
• audit-ux-and-save-files — usability audit from real personas walking their journeys, per-issue findings to ux-findings/[yy-mm-dd]/..., ending with a prioritized recommendations report (reports, doesn't fix).

/plugin install yk-frontend@yigitkonur

📝 review & completion

judge a change for merge-readiness, triage feedback, and verify "done".

• run-review — one entry point, four modes: (a) do a pr/branch review, (b) open your branch as a self-review pr, (c) triage received feedback, (d) delegate to codex review.
• run-codex-review-loop — multi-lens or multi-branch codex review loops; independently verify findings and optionally fix confirmed issues in isolated worktrees until convergence.
• audit-completion — audit task / session / plan / branch completion claims with evidence; remediate to terminal status.

/plugin install yk-review@yigitkonur

🔬 research & discovery

answer questions and find things with real web evidence. ships the internet-researcher-* subagents.

• run-research — one technical question, current web + reddit practitioner evidence, source-backed synthesis, optionally fanned across subagents.
• run-deep-research — wave-based multi-file corpus research over 5+ entities or a market/category; parallel subagent orchestration and evaluation.
• run-github-scout — adaptive github repo discovery, shortlisting for a concrete need, oss comparison with repo evidence.

/plugin install yk-research@yigitkonur · agents only: /plugin install yk-researchers@yigitkonur

🤖 live automation

drive a browser, a phone, or a terminal workspace mid-session.

• herdr — control herdr panes, tabs, workspaces, worktrees, commands, and coding agents without stealing focus.
• mobilerun-control — drive a connected android phone via the mobilerun cli: tap/type/swipe/read by box-center, deterministic multi-step on-device tasks.
• project-manager — supervise coding agents with adversarial verification, automated monitoring, structured progress reports, and Herdr-native pane coordination.
• run-agent-browser — agent-browser cli: @ref snapshots, sessions, forms, extraction, screenshots, headed/stealth, provider runs.
• run-agent-device — agent-device cli for ios app testing: settle-first snapshot/press/fill loop, evidence capture, cross-layer bug triage, runtime-freshness + fresh-state discipline, fix-and-retest.
• use-chatgpt-by-applescript — drive macOS ChatGPT desktop app via AppleScript or remote SSH: automated web research (@Browser), computer actions (@Computer), instant 0.3s AX status monitoring, and clipboard-free Markdown AST extraction.

/plugin install yk-automation@yigitkonur

🚀 ops & release

deploy, maintain, offload, publish.

• use-cloudflare-tunnel — expose localhost ports and multi-service apps (frontend SPA + backend APIs) to the internet via Cloudflare Tunnel: quick tunnels, named zero trust tunnels, same-origin unified proxy to eliminate mixed-content blocks, ego-browser & webhook testing.
• use-sentry — fullstack Sentry observability & operations: zero-to-one setup with research-mcp (max 20 keywords), 4-pillar enterprise feature audit, envelope tunneling for ISP DNS sinkholes, and token-efficient 4-rung CLI triage.
• run-railway — railway cli: deploys, logs, env vars, link, ssh, db shells, scaling, installed-vs-docs version-drift routing.
• deploy-coolify-cloud — deploy/update docker-compose services on coolify cloud via the api: verified create/patch/urls-domain/env-var contracts, base64 compose, custom domains + TLS, cross-service networking, and box-level deploy verification.
• ci-cd-optimize — diagnose or optimize slow CI/CD by measured bottleneck — GitHub Actions, GitLab CI, CircleCI, Buildkite, monorepos, Docker builds, runner queues, deployment paths, and Swift/Xcode CI — while preserving required checks, cache correctness, and exact-artifact verification.
• run-repo-cleanup — finish a project: review + merge every live branch and worktree into main locally (no prs), retire dangling branches, sweep junk to a gitignored trash.
• run-ts-cleanup — clean up a typescript codebase: dead code, unused deps, ai slop, weak types, tangled structure — knip + biome/oxlint/eslint + tsc, in reversible waves.
• publish-npm-package — npm releases via github actions: trusted publishing, NPM_TOKEN, provenance, semantic-release, changesets, release-please.

/plugin install yk-ops@yigitkonur

✍️ writing & editing

rewrite supplied content for real readers while preserving evidence, locale, and document structure.

• convert-to-natural-writing — diagnose, rewrite, and publication-review robotic, generic, or AI-sounding multilingual copy in plain text, Markdown, MDX, or HTML; preserves claims and structure without detector theater or invented personality.

/plugin install yk-writing@yigitkonur

🧩 skill authoring

build and harden skills themselves.

• build-skill — create/redesign/merge a skill with evidence-based research and comparison before writing skill.md.
• audit-skill-by-derailment — stress-test an existing skill.md by running a fresh subagent on a real task and editing the skill where the trace shows friction.

/plugin install yk-skills@yigitkonur

notes

• every enabled skill costs context. that's the whole point of shipping per-skill Codex and Claude plugins, plus small Claude bundles, instead of one blob — enable what you use.
• the plugin paths and the skills cli read the canonical skills/ files. generated metadata lives in .claude-plugin/marketplace.json, .codex-plugin/plugin.json, .agents/plugins/marketplace.json, and the self-contained Codex packages in plugins/; the version comes from VERSION and ci bumps it on every push to main.
• naming/taxonomy → NAMING.md. structure + contribution checklist → CONTRIBUTING.md.
• spec: agentskills.io/specification. plugin/marketplace docs: code.claude.com/docs/en/plugin-marketplaces.
• duplicate skill triggers? known claude code thing (#27721).

contributing

see CONTRIBUTING.md. new skills use a verb from the 12-verb registry, pass python3 scripts/validate-skills.py, and land in the marketplace via python3 scripts/gen-marketplace.py.

license

mit