Nine Free AI Agent Skills Worth Installing Right Now

An AI skill is a reusable recipe card that teaches an assistant how to do a recurring job; some skills are plain written instructions while others bundle scripts, reference files, and ready-made commands. The presenter notes these recipes work across ChatGPT Work, Codex, Claude Code, Co-work, and similar agent harnesses. Underneath sits an open standard: every skill folder starts with a SKILL.md entry file, and its name plus description fields decide when the assistant reaches for it.

Install flow: one link, one sentence

Installing one is strikingly simple: copy the skill link from its GitHub repository, paste it into the assistant, and ask it to follow the setup instructions, with the whole thing done in about a minute. To call a skill you press @ and pick it from the list, though most tools also trigger the right skill on their own from context. Per the OpenAI help page, skills inside ChatGPT are reusable, shareable workflows that fire automatically, so each installed skill quietly joins everyday chats without extra settings.

A research skill that reads the room

The first pick is Last 30 Days , a research tool that finds what people actually say about a topic right now; instead of leaning on classic web search it scans community sources such as Reddit, X, YouTube, TikTok, Instagram, Hacker News, and Polymarket, then compresses everything into a short readable brief. The last30days entry on AgentSkillsHub confirms the Reddit and Hacker News heavy coverage, the Python build, and the free license. The host tests it on Jev, the new tool from TypeSafe; a five-minute run over 14 Reddit threads and 13 Hacker News items returns a balanced digest.

Jev in the TypeSafe docs explains why that demo matters: the model returns typed answers with confidence scores instead of free text, so the zero-hallucination claim means it never invents a category, not that it never errs. The author then states the central thesis: skills are starting points, not finished products. He has Codex run the skill, pour the output into his own site with use cases and contrarian takes, and finally packages everything as an extended new skill. The message is plain: whoever adapts a ready-made skill to their own setup gets the most value.

Two rivals that make video from code

For video work two skills go head to head: Remotion and Hyperframes . Per the Remotion docs, setup is a single command and the assistant opens a full editing studio built on React code; the requested skills-themed intro animation comes out nearly flawless on the first try. The HyperFrames site quickstart leans on an engine that deterministically compiles an HTML file into MP4, and the same prompt impresses there too. The gap shows in a 15-second Utah national parks road-trip animation: Remotion follows the roads correctly but adds confusing spins, while Hyperframes wins on aesthetics with a better camera angle and closing zoom-out. The verdict is balanced: Remotion shines for abstract intros, Hyperframes for maps and camera moves, so install both and pick per job.

Security scanning and a human touch

The risky side of installing skills is covered by SkillSpector in the Nvidia docs; the scanner checks skill folders against 68 vulnerability patterns in 17 categories and reports prompt injection, data exfiltration, and excessive permissions before anything is installed. Research quoted on the project GitHub page finds over a quarter of skills carry flaws, which is why the host scans every foreign skill as a habit. A clean report never equals absolute safety, yet combined with personal judgment it adds a useful extra layer. The same rigor continues on the writing side with the Humanizer skill: the assistant first drafts three promo posts about Jev, and the familiar artificial tics plus hyped closings are shown one by one.

The Humanizer call scrubs those tells and leaves copy that reads as if a human wrote it; the short samples fill the screen, though the technique pays off most in longer pieces.

The video-watching skill closes a blind spot of chatbots: given an Instagram camping-meal clip whose caption lacks the method, it recovers the full recipe from on-screen text and timings. The skill-free attempt at the same link hits a wall, and uploading the file works only after extra back-and-forth rounds. In short the skill bundles frame extraction and narration into one flow and even downloads the clip to process it alone when a site blocks viewing. For builders, the Superpowers pack steps in: obra/superpowers on GitHub gives the assistant a development discipline through 14 workflow skills like brainstorming, plan writing, test-driven development, and systematic debugging.

Garry Tan's piece in the Y Combinator library describes the sibling project Gstack; inside that 23-tool set the office-hours skill that stress-tests a product idea with six forcing questions and the browser-driving QA watcher stand out (ycombinator.com). The closer belongs to the Impeccable design skill; the 23 commands on Impeccable.style run from making dull designs bolder to adding meaningful motion, fixing typography, and full critique. The author's one-prompt Utah parks passport site starts clean yet bland; the bolder command revives it with dark green sections and strong type, while the delight command adds a passport-stamp animation as a small moment of joy. The critique command then lists concrete notes on design, function, and mobile fit, proving the skill is a working tool rather than decoration.

The practical bottom line

The big picture is simple: a few well-chosen skills turn an ordinary assistant into a reliable coworker across research, production, and design. The same logic extends to the free 12-pack of conversational skills mentioned along the way; one-command helpers like brand voice, a custom content engine, and a video launch kit wait to strengthen everyday writing. Start with two or three, scan what comes from strangers, adapt the output to your own setup, and keep only the skills you use every week.