The best agent skills, according to hackers
23 agent skills recommended in Hacker News comments over the past year, ranked by how many separate people recommend each one.
- Most recommendedSuperpowers138 people recommend it
- Least regrethandoff skill for Claude Code17 recommend, 0 criticize
All agent skills
- SuperpowersA free Claude Code skills plugin that makes the agent ask clarifying questions and write detailed plans before coding; critics say it burns tokens and forces unneeded TDD.138recommend
- Claude SkillsReusable markdown instruction packs that Claude loads only when relevant, so it follows project procedures without bloating context. Critics say triggering is unreliable and it locks users into Anthropic.59recommend
- grill-me skill by Matt PocockMakes the AI question your plan until you both agree on the design, so it builds what you expect. Critics say it still scope creeps and can produce weak code.39recommend
- Get Shit DoneFree prompting and workflow system for Claude Code that plans work, keeps context small and lands features on larger projects. Critics say it is heavy on turns, tokens and up-front planning.31recommend
- Matt Pocock's skillsRecommended for well-tuned agent skills such as grilling, TDD and planning that produce higher-quality code with coding agents. Caveats: some find it excessive and the writing style hard to follow.27recommend
- Claude Code skillsPeople recommend Claude Code skills for packaging repeatable workflows into reusable commands that load only when needed. The most common caveat is that Claude sometimes fails to activate them.21recommend
- PonytailRecommended as a plugin that keeps AI coding agents terse, with smaller edits and less code. Main caveat: results are hit or miss and depend heavily on context.20recommend
- frontend-design skill for Claude CodeAnthropic's official Claude Code skill that makes Claude build front ends beyond generic AI looks; common caveat is last-mile debugging and templated-looking results.17recommend
- handoff skill for Claude CodeLets you pick up a Claude Code session in a new session or another AI tool like Codex, and people find it more useful than /compact.17recommend
- Axiom by Charles WiltgenFree open-source Claude skills and agents that help coding assistants build and audit iOS and macOS apps; its main caveat is that it does little for code review or generation.16recommend
- GSD (Get Shit Done)People recommend it for running multi-step AI coding work in phases with fresh context, so the context never needs compacting. Critics say it uses many tokens and suits only large projects.15recommend
- ImpeccablePeople recommend it for critiquing designs, spotting design slop, and steering AI-generated frontend work toward better design. Critics say it leaves the core design largely unchanged unless prompts are very explicit.15recommend
- gstack by Garry TanPeople recommend it as a set of Claude Code skills and browser tooling that speeds up development and improves code quality, though critics say it is token-heavy and overhyped.14recommend
- Claude Code skills for building Godot gamesRecommended as a base for prototyping Godot games with Claude Code. Critics say the generated games are incomplete and the title overstates them.14recommend
- Humanizer by bladerPeople recommend it for stripping common AI writing tells like 'not x, but y' and bolded bullets. Critics say the output still reads as AI-written.13recommend
- Claude Code ReviewRecommended as a convenient way to get AI code review inside a Claude Code session that finds good issues. Critics say it has blind spots and costs $15-25 per review.11recommend
- Universal Claude.mdInstruction set that measurably cuts Claude's output verbosity, though critics say it may hurt quality on complex tasks and is not benchmarked for accuracy.10recommend
- Claude Code /simplifyClaude Code skill that trims over-engineering and verbosity in code; people find it effective, though some engineers get weaker results and a benchmark found no test pass gains.9recommend
- I-have-ADHD skillPeople recommend it for cutting verbose, wall-of-text answers from Claude so the point comes first. Critics say the effect fades over long sessions.9recommend
- Superpowers brainstorming skillPeople recommend it because the AI asks questions before building, giving a consistent process and high-quality code. Critics ask whether it beats a vague prompt.8recommend
- skill-creatorRecommended for generating new agent skills, including subagent tests of whether a skill helps. Critics say it micromanages models and works best as a scaffold for simple skills.8recommend
- Wayfinder skill for ClaudeA Claude skill that clears the path through unclear work before coding starts, switching to questioning mode when needed and suggesting direct work for light tasks. Some find it most useful for big projects.5recommend
- SimpleEnglish skill by AminBlgFree agent skill that makes LLM-written code comments and docstrings clearer, based on the ASD-STE100 simplified English standard. No criticism recorded.5recommend