Skills & plugins
Addy Osmani Agent Skills
NEW + TRENDING + HOT: Production-grade engineering skills for AI coding agents from Addy Osmani, the engineering lead running Chrome's developer experience at Google. Repo just hit 102K+ GitHub stars (currently sitting at the top of GitHub Trending), with 15 releases shipped in 2026 and v0.6.13 dropping Oct 13, 2026 — four days ago. Cutting-edge angle: it encodes the SDLC work senior engineers do but agents skip by default (spec, plan, build, test, review, ship) as 25 installable skills plus 9 slash commands (`/spec`, `/plan`, `/build`, `/test`, `/constraints`, `/review`, `/webperf`, `/code-simplify`, `/ship`). Each skill is a workflow with checkpoints that produce evidence (write the failing test first, run it, watch it fail, write the minimum code to pass, watch it pass, refactor) instead of reference docs the agent can ignore — the same evidence-producing pattern Google engineering practices are built on. Killer features for vibe coders: (1) auto-activation — designing an API triggers `api-and-interface-design`, building UI triggers `frontend-ui-engineering`, no manual command needed; (2) `/build` is one-shot — it generates the plan and implements every task in a single approved pass, still test-driven and committed individually, but removes the human stepping between tasks; (3) cross-harness via the open `skills` CLI — `npx skills add addyosmani/agent-skills` installs all 25 into 70+ agents (Claude Code, Cursor, Codex, Copilot, Cline, OpenCode, Hermes, Antigravity, Zed, Gemini, and more), with per-skill installs for the headline skills (`code-review-and-quality`, `test-driven-development`, `interview-me`); (4) `/review` overrides Claude Code's built-in to enforce a five-axis review before merge; (5) `/interview-me` does one-question-at-a-time requirements interrogation so the agent ships what you meant, not what you said. Strongest social proof on the resource page: 102K+ stars (gained ~75K in the last 60 days alone), Global Rank #118 on star-history.com, 52K milestone with `#1 trending repo` callout from Addy on LinkedIn, repeated returns to GitHub trending through summer 2026, currently at the top again, written by one of the most recognized engineering-lead voices in the industry and recommended alongside `obra/superpowers` and `ponytail` in every serious Claude Skills roundup. Direct competitor: `anthropics/skills` — but Addy's pack is opinionated, SDLC-shaped, and built by a senior engineer for senior engineers, while the official Anthropic one is broader and more generic. Category: `skills-plugins` because the killer feature is treating an agent harness like a buildable engineering org rather than a chat box. Vibe coders shipping production agents will recognize it as the missing piece that makes Claude Code and Cursor finally behave like a tech lead.
Documentation reviewed 2026-10-17
Skills & plugins
Cloudflare Security Audit Skill
Hot-off-the-press (launched Sept 17, 2026 alongside Cloudflare's "Build your own vulnerability harness" post) MIT-licensed coding-agent skill from Cloudflare that turns Claude Code, Codex, OpenCode, Cursor, and similar agents into structured security auditors. Instead of one agent hallucinating vulnerabilities, it runs a six-phase adversarial pipeline inside a single repo: three parallel research agents map architecture and trust boundaries; per-class Hunter agents try to break code rather than review it; adversarial validators try to disprove each candidate; a fresh agent independently re-verifies every surviving finding against the source; final reports land in REPORT.md plus machine-readable findings.json against a strict schema, with zero-dep validators. Model-agnostic on purpose: Cloudflare swaps the frontier model in the harness because different models surface different bugs, exactly the architecture vibe coders need to ship agent-built apps without shipping a CVE. Adoption signal is wild for a four-day-old repo: 18K+ GitHub stars, ~2.4K stars in the last 24h alone, hitting the GitHub trending top 3, covered by Korben and Remio, and the same skill Cloudflare itself uses internally as the seed for its fleet-wide vulnerability discovery harness. Install with `npx skills add https://github.com/cloudflare/security-audit-skill --skill security-audit` and run it against any vibe-coded repo before you merge.
Independent review pending
Skills & plugins
DESIGN.md
Google Labs' new spec for vibe coding with consistent UI.
Independent review pending
Skills & plugins
gstack (by Garry Tan)
Use Garry Tan's exact Claude Code vibe coding setup.
Independent review pending
Skills & plugins
last30days-skill
The #1 GitHub trending AI agent skill of early April 2026 (19.6K+ stars in days).
Independent review pending
Skills & plugins
Pi Coding Agent
A genuinely fresh take on vibe coding: instead of shipping one fixed agent workflow, Pi is a minimal terminal harness you reshape with extensions, skills, prompt templates, themes, and npm-distributed packages.
Independent review pending
Skills & plugins
PI-mono
Agent toolkit for vibe coders who want full control: coding-agent CLI, unified multi-model LLM API, TUI/web UI packages, and extensible skills/subagents architecture.
Independent review pending
Skills & plugins
Ponytail
One of the most genuinely fresh and most-loved launches for vibe coders of September 2026. Open-source YAGNI plugin for Claude Code, Codex, OpenCode, Hermes Agent, Devin CLI, GitHub Copilot CLI, and Gemini CLI that makes coding agents write the least code that works. Before adding anything, the agent checks whether the change is even needed, whether the repo already does it, whether the standard library or a native platform API does it for free, and whether an existing dependency already covers it. Published benchmarks show ~54% less code (up to 94% on individual tasks), ~20% cheaper runs, and ~27% faster completion. Took Product Hunt Launch of the Day on Sept 5, 2026, hit 44K GitHub stars in its first nine days (now 129.7K+), trended #2 on GitHub, sparked HN and Reddit debates, and v4.9.0 ships fast across 20 agent surfaces.
Independent review pending
Skills & plugins
Superpowers
TDD-enforced skills framework that makes AI coding agents follow disciplined development methodology.
Independent review pending
