ScrapeCreators social-media-research-skills: agent skills for outlier posts and comment mining

What the skills actually solve for social research workflows

Asking a coding agent to "analyze this creator" usually produces a plausible essay with no data behind it. This repository tries to fix that by shipping named workflows that carry their own method: what to fetch, how to judge it, and what artifact to emit. The README frames the distinction directly, saying this is "not just endpoint routing" and that the goal is to give an agent "complete research workflows." Twelve workflow skills are listed, from outlier-post-finder and comment-mining through ad-library-teardown, trend-discovery, influencer-prospecting and content-repurposing.

The audience is narrower than the topic list suggests. These are for growth marketers, content strategists, and analyst teams who already run Claude Code, Cursor, Codex, Gemini CLI or another Agent Skills host and who want the agent to produce a brief, a table or a CSV rather than raw JSON. A developer who just wants to call an endpoint once does not need a skill; the scrapecreators-api skill covers that case and is described as the data layer the other workflows call into.

How the skills layer sits on top of the ScrapeCreators API

The repository layout is flat and readable: a skills/ directory, a scripts/ directory, an AGENTS.md and a CLAUDE.md at the root, and a .claude-plugin/ folder. The README's diagram places scrapecreators-api at the top, feeding the twelve workflow skills underneath. That ordering is the design: a workflow skill decides what evidence it needs, then defers to scrapecreators-api for endpoint choice and pagination guidance.

Six design principles are listed in the README, and two of them change how output should be read. The first is baseline-aware analysis, which says performance should be judged against a creator's own normal performance rather than raw vanity metrics. That is what makes outlier-post-finder meaningful: an outlier is defined relative to a creator's baseline, not by an absolute view count. The second is public-data only, with the README stating that ScrapeCreators extracts public social data and that skills should not promise logged-in or private data. The remaining principles cover cited source URLs, verbatim quotes, and ending with patterns or test ideas instead of a wall of JSON.

Platform breadth is where the marketing claim and the skill table diverge slightly. The introduction names TikTok, Instagram, YouTube, Reddit, X/Twitter, LinkedIn, Facebook, Threads, Bluesky, Pinterest, Rumble and ad libraries. Individual skills are more specific: transcript-intelligence covers TikTok, Instagram, YouTube, Facebook, X, LinkedIn, Rumble and Reddit, while ad-library-teardown is scoped to Meta, Google and LinkedIn ads. Treat the long list as coverage of the underlying API, not as a promise that every skill works on every network.

Installing the skills and running a first outlier search

Installation goes through the skills CLI rather than pip, even though the repository's primary language is Python. The README gives a single command:

npx skills add ScrapeCreators/social-media-research-skills

The README states this works with Claude Code, Cursor, OpenAI Codex, GitHub Copilot, Gemini CLI, Windsurf, VS Code and other agents that follow the Agent Skills spec. After running it, the skills should appear in your agent's skill list; the repository does not document a verification command, so check your host's own skill listing.

Next, export the API key. The README shows the variable name and a placeholder value:

export SCRAPECREATORS_API_KEY=sk_...

A key is issued at scrapecreators.com, per the README. Nothing in the repository documents key rotation, scoping or expiry, so treat the environment variable as the single point of configuration.

With both in place, invoke a skill by asking for the outcome, not the endpoint. The README's own example prompt is:

Find the outlier posts for @starterstory on YouTube Shorts from the latest page of videos.

Expect an outlier table with the posts that beat the creator's baseline, plus the repeatable patterns and hooks the skill extracted. If the agent returns raw JSON or an endpoint listing instead, it routed to scrapecreators-api rather than the workflow skill; restate the request as a business outcome.

Where these skills are the wrong tool

The public-data constraint is the first hard boundary. If your question depends on private groups, logged-in feeds, DMs or gated analytics, this project is the wrong layer, and the README says so explicitly by warning against promising logged-in or private data.

The second boundary is the API key. Every workflow skill ultimately calls ScrapeCreators, so the skills are only as useful as your plan and quota allow. The repository contains no local scraper, no caching layer and no fallback data source. If the key is missing or the quota is exhausted, the skills have nothing to reason over. That also means cost scales with research volume in a way the README does not quantify; there is no rate-limit or pricing table in the repository.

The third boundary is that skills are prompts plus method, not deterministic software. Two runs of comment-mining on the same post can produce differently grouped themes, because the grouping is done by the model following the skill's instructions. The README's cited-outputs principle pushes toward source URLs and verbatim quotes, which helps you audit a result, but nothing in the repository guarantees reproducible categorization. If you need a fixed taxonomy with stable counts, build that pipeline yourself and use the API directly.

Finally, sentiment is treated with caution rather than as a headline metric. The social-listening-brief skill is described as producing "sentiment caveats" alongside themes and cited examples, which is a more honest framing than most listening tools offer, but it also means you should not expect a clean sentiment score to drop into a dashboard.

How this differs from calling the ScrapeCreators API or a generic prompt

The obvious alternative is skipping the skills and calling the ScrapeCreators API from your own script, using the docs at docs.scrapecreators.com. The difference is where the judgment lives. A direct API call returns posts or comments; you still have to decide what counts as an outlier, which comments are objections versus praise, and how to present the result. The skills encode those decisions as instructions, and the README's design principles make the encoding explicit: baseline-aware comparison, verbatim language, cited sources, actionable endings.

A second alternative is a general-purpose research prompt with no skill installed. That works until the agent needs endpoint details it does not have; the repository's answer is the scrapecreators-api skill, which the README says the workflow skills should consult as the data layer. The trade-off is coupling: adopting these skills means adopting ScrapeCreators as the data source, and the MIT licence on the skill files does not extend to the API behind them.

There is also an MCP integration documented at docs.scrapecreators.com/integrations/mcp. If your agent already talks to ScrapeCreators over MCP, the marginal value of the skills is the workflow method rather than the connectivity, and the two can coexist.

Maintenance, licence and what an upgrade costs you

The repository is not archived, and the last push was on 2026-08-26, which is recent enough that the skill set should track current endpoint behaviour. There are no retrieved releases, so versioning appears to follow the main branch rather than tagged versions. That matters for upgrades: reinstalling with npx skills add ScrapeCreators/social-media-research-skills pulls whatever is on main, and the README does not document a rollback path or a pinned version. If you need reproducibility, record the commit you installed from yourself.

The licence is MIT, which covers the skill files, the scripts directory and the plugin manifest. It does not cover the ScrapeCreators service, the API key, or the data you retrieve; those are governed by your account terms at scrapecreators.com, and the repository does not restate them. The practical implication is that you can fork and edit the skills freely, which is likely the intended workflow given that skills are markdown instructions, but you cannot fork the data layer.

Upgrade cost is mostly re-reading the skill you depend on. Because skills are instructions rather than a library with a stable interface, a change in wording can change output shape, and nothing in the repository signals breaking changes. Teams running these in production should diff the skill files after each reinstall.

Editorial conclusion

Adopt it if your team already pays for ScrapeCreators and wants repeatable research artifacts instead of one-off prompts; the skills are thin markdown workflows over a paid data layer, so the API key is the real dependency. Do not adopt it if you need private or logged-in data, or if you want a self-hosted scraper, because the README states the project extracts public data only. Before committing, verify that npx skills add ScrapeCreators/social-media-research-skills detects your agent, and confirm your ScrapeCreators plan covers the endpoints a skill such as ad-library-teardown actually calls.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. Project website
4. README
5. ScrapeCreators/social-media-research-skills on GitHub