apify/agent-skills: Apify's Skill Pack for Coding Agents, Read Against Its Own README
Collection of Apify agent skills
At a glance
- What is it?
- Apify ships five agent skills plus a slash-command pack that teach coding agents to pick Apify Actors, scaffold new ones, and wrap existing code as Actors. The design is a thin instruction layer over a paid platform, and the README is more specific about prompts than about cost.
- Who is it for?
- Adopt apify/agent-skills if you already run Apify Actors and want an agent to choose and chain them from a plain-language prompt, or if you are building an Actor and want scaffolding help. Do not adopt it if you need per-run cost predictability or you are not on the Apify platform at all, because every skill points back at Apify Actors and the Apify Store.
- Can I use it commercially?
- Not without permission. GitHub finds no licence file in the repository, and without a licence all rights are reserved by default: you may read the code but not reuse it. Check the README, or ask the authors, before using it.
- Is it still maintained?
- Yes. The repository last received commits 6 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The gap between a coding agent and a scraper it has never seen
A coding agent knows how to write a requests loop. It does not know which Apify Actor handles Google Maps reviews, what input shape that Actor expects, or how to read the run's dataset afterwards. That gap is what this repository fills. Each skill is a folder of instructions and reference material that a host agent (the README names Claude Code, Cursor, Windsurf, Codex and Gemini CLI) loads before it answers a scraping question. The audience is narrow on purpose: developers who already have an Apify account and want an agent to drive the platform, plus developers building or porting Actors who want the SDK patterns and schema conventions in context. Anyone outside the Apify ecosystem gets nothing from it, because every skill resolves to an Apify Actor run.
Five skills, one of which does most of the work
The README lists five skills. apify-ultimate-scraper is the one most users will install first: it carries a curated set of 130+ Actors across Instagram, Facebook, TikTok, YouTube, X, LinkedIn, Reddit, Google Maps, Google Search, Google Trends, Amazon, Walmart, eBay, Booking.com, TripAdvisor, Airbnb, Yelp, Telegram, Snapchat and GitHub, and the README states it falls back to searching the full Apify Store for platforms not in that list. The other four are builder-side: apify-actor-development for creating, debugging and deploying Actors in JavaScript, TypeScript or Python; apify-actorization for wrapping existing code (JS/TS SDK, Python async context manager, or a generic CLI wrapper for other languages); apify-generate-output-schema for deriving dataset_schema.json, output_schema.json and key_value_store_schema.json from existing source; and apify-integration-development for building an official integration into someone else's product, with per-category capability matrices and definition-of-done checklists. A separate commands/ pack adds slash commands such as /create-actor. The split matters: only the first skill is for data consumers, and the README's example prompts are almost all about that first skill.
Installation is one npx command, and the README cuts off before showing the rest
The documented path is `npx skills add https://github.com/apify/agent-skills --skill apify-ultimate-scraper`, run from a shell, after which the agent is expected to handle Actor selection, input shaping, run management and result formatting. The README's Quick start section then shows a natural-language prompt example about scraping Google Maps results to CSV. The Installation section begins the Claude Code path with `/plugin m` and the supplied text ends there, so the remaining plugin commands and any per-editor instructions for Cursor, Windsurf, Codex or Gemini CLI are not visible in this material. Treat the skills.sh homepage (https://skills.sh/apify/agent-skills) as the place to look for the install variants the README does not show. No configuration keys appear in the excerpt either, which means the question of which Apify token the agent uses and where it is stored is unanswered here. That is the first thing to check before running anything against a real account.
What the skills actually contain, and what the README does not say about it
The README describes skills as directories (skills/apify-ultimate-scraper/, skills/apify-actor-development/ and so on) and says the Actor development skill bundles references covering actor.json, input, output, dataset and key-value schemas, logging and standby mode. That is a description of reference documents, not executable code. The practical consequence is that behaviour depends on the host agent: the same skill folder read by two different agents can produce different Actor choices and different input JSON. The README does not state a minimum agent version, a token budget, or how the fallback to the full Apify Store search is bounded. If you need deterministic Actor selection, this is the wrong layer, because the selection step is a model decision, not a lookup table you control.
The cost model is the limitation the README does not address
Every skill here ends in an Apify Actor run, and Actor runs consume Apify platform credits or paid usage. The README does not mention pricing, free tiers, or how an agent should estimate the cost of a scrape before starting it. That is a real failure mode in agent workflows: a prompt like the lead-generation example (find restaurants on Google Maps, scrape reviews, crawl their websites for emails, export a ranked CSV) chains several Actors over an unknown number of pages, and nothing in the skill description says the agent will ask before spending. If you are running this against a production account with a shared budget, you need your own guardrail outside the skill. A second limitation is scope: the integration-development skill targets building an Apify integration into another product, so it is irrelevant if you just want data out.
When a direct apify-client call beats an agent skill
The README's own integration-development skill names apify-client and the REST API as the direct-client route, which makes the comparison concrete. A scheduled nightly job that pulls the same Actor with the same input does not need an agent in the loop: an apify-client call with a fixed actor ID and a fixed input object is cheaper, faster and reproducible, and it fails in a way you can log. The skill layer earns its place when the Actor is not known in advance, when the input depends on a question phrased in natural language, or when a task spans several platforms and the chaining decision is the hard part. For the market-research example (pricing and review counts across Amazon and Walmart), the agent skill is doing real work. For a cron job, it is overhead.
Maintenance, licence and the Apache 2.0 badge
The repository metadata supplied here reports no licence, while the README carries an Apache 2.0 badge linking to a LICENSE file. Those two signals disagree, and the badge is the weaker one: check the LICENSE file in the repository before you depend on the terms, and read it as a question for your own legal review rather than something this article can settle. On maintenance, the last push recorded is 2026-09-09 and no releases were retrieved, so there is no versioned artefact to pin. That matters more than usual here because the skills are instruction files consumed at install time: a change to a skill folder changes agent behaviour with no version bump to notice. If you vendor the skill into your own repository, you get a diff to review. If you install it fresh each time, you do not. The README also points to apify/awesome-skills for community-built, domain-specific skills, which is where to look if the five official skills do not cover your domain.
Editorial conclusion
Adopt apify/agent-skills if you already run Apify Actors and want an agent to choose and chain them from a plain-language prompt, or if you are building an Actor and want scaffolding help. Do not adopt it if you need per-run cost predictability or you are not on the Apify platform at all, because every skill points back at Apify Actors and the Apify Store. Before installing, check the LICENSE file for the exact terms, since the repository metadata reports no licence while the README badge says Apache 2.0, and confirm what a run costs on your plan.
Community notes