blacktwist/social-media-skills: Prompt Packs for Social Writing Agents
AI agent skills for social media content strategy, creation, and analysis across text-first platforms
At a glance
- What is it?
- This repository ships fifteen Markdown skill definitions that Claude Code, Cursor and other agents load to write and analyse social posts. The interesting part is not the prompts, it is the context file every other skill reads from, and the MCP fallback that decides whether the agent publishes or just drafts.
- Who is it for?
- Adopt it if you already run Claude Code or Cursor and want your social writing to inherit a fixed voice and audience definition instead of re-explaining yourself in every prompt. Skip it if you need an autonomous scheduler: without the BlackTwist MCP server connected, every skill drops to advisory mode and you post by hand.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 138 days ago.
- What is it written in?
- Mainly Shell, 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
What the skill catalog actually covers
The repository is a catalog of fifteen skills split into four groups. Foundation holds one entry, social-media-context-sms, described as capturing platform context, audience details, content pillars and tone preferences, and the README states it is used by all other skills. Strategy covers content pillars and positioning (content-strategy-sms), publishing cadence and scheduling (content-calendar-sms), and per-platform tailoring (platform-strategy-sms). Creation covers single posts, multi-post threads, LinkedIn carousels, captions for visual-first platforms, repurposing, and hook writing. Analysis covers metric interpretation, follower trend tracking, content pattern detection, and an optimisation advisor that turns performance data into specific edits.
The split matters because it maps to how people actually work. Writing a hook is a different task from deciding which topics deserve a hook, and the repository keeps those in separate files so an agent loads one without the other. The target user is someone already running an agent in a code or writing environment who wants social output to follow a defined voice rather than whatever the model defaults to. It is not a scheduling product and it does not host anything.
The context file is the load-bearing part
Every skill in the Strategy, Creation and Analysis groups is described as drawing on social-media-context-sms. That makes the foundation skill the real mechanism: you fill in platform context, audience, pillars and tone once, and the downstream skills read those values instead of guessing. An agent asked to write a thread will still produce a thread without it, but the output will drift toward generic LinkedIn voice, because nothing in the prompt pins the audience or the house style.
This is also the design's main cost. A single shared context file means a change to your tone preferences propagates everywhere at once, which is convenient right up to the moment you want one platform to sound different. The README lists platform-strategy-sms as the place for per-platform approach, so the intended pattern is a shared core plus a platform layer, but the material does not show how conflicts between the two are resolved. Treat that as an open question and test it with your own context before trusting it across four platforms.
Installation paths and what lands on disk
The README gives six options. The recommended one uses the npx skills CLI:
npx skills add blacktwist/social-media-skills
Selective install takes a skill list:
npx skills add blacktwist/social-media-skills --skill post-writer-sms hook-writer-sms
and npx skills add blacktwist/social-media-skills --list enumerates what is available. According to the README, this writes into .agents/skills/ and symlinks into .claude/skills/ for Claude Code compatibility. If you prefer not to run the CLI, a clone plus cp -r social-media-skills/skills/* .agents/skills/ does the same thing by hand, and git submodule add https://github.com/blacktwist/social-media-skills.git .agents/social-media-skills keeps updates tied to a commit you control. There is also a SkillKit route (npx skillkit install blacktwist/social-media-skills) for spreading the same skills across Claude Code, Cursor and Copilot, and a manual Cowork path that uploads one zip per skill through Cowork > Customize > Skills.
Pick the submodule or the fork if you plan to edit the prompts, because both keep your changes separate from upstream. The CLI path is the least friction but overwrites on update.
MCP integration decides whether it publishes or advises
The README names BlackTwist as the recommended MCP integration. When that server is connected, the skills use it directly to publish and schedule posts, fetch analytics and engagement data, manage drafts and queued content, and track follower growth. When it is not connected, the README says skills fall back to advisory mode: they generate content and instructions for manual posting.
That fallback is the honest boundary of the project. Without the MCP server, the Analysis group has nothing to analyse unless you paste metrics in yourself, and the Creation group produces text you copy into a browser tab. The skills still work, but the value drops to prompt scaffolding. The README points to tools/REGISTRY.md for the full tool reference, which is where you would check exactly which tools each skill expects before assuming a given skill runs unattended.
Where this is the wrong tool
The repository covers text-first platforms in depth (LinkedIn, Twitter/X, Threads, Bluesky) and visual-first platforms only through caption-writer-sms, which the README describes as writing captions for Facebook, Instagram, TikTok and Pinterest. If your work is video editing, thumbnail design or Reels production, this gives you the words around the asset and nothing else. The catalog does not claim otherwise, but the platform list can read as broader coverage than it is.
A second limit is measurement. The Analysis skills interpret metrics, and the README does not describe how those metrics arrive beyond the BlackTwist MCP path. If you publish through a different scheduler, you are feeding numbers in by hand, and the pattern analysers are only as good as what you paste. A third is that these are prompt files: nothing enforces that the agent follows them. A model that ignores the tone section produces off-brand copy, and the fix is prompt editing rather than a config flag.
How it differs from a general-purpose prompt library
The obvious alternative is a folder of your own prompts, or a generic prompt collection for social writing. The difference is structure. A personal prompt file tends to be one long instruction that covers voice, format and platform at once, so changing the audience means rewriting the whole thing. This repository separates the shared context from the task, which means the audience definition lives in one place and the hook-writing instructions live in another.
Compared with a scheduling tool that bundles AI caption generation, the split runs the other way. A scheduler owns publishing and treats the writing as a feature inside it. Here the writing is the product and publishing is delegated to an MCP server, which is why the project degrades to advisory mode so cleanly. If you want one dashboard that drafts and posts, this is more assembly than you asked for. If you want the writing layer to be portable across agents, the separation is the point.
Maintenance cost and the MIT licence
The release history shows v1.1.0 in April 2026, v1.2.0 later that month and v1.3.0 on 2026-05-01, so the catalog is moving at roughly a minor release every two to three weeks. That cadence cuts both ways: new skills and prompt fixes arrive, and if you forked or copied the files, you re-merge by hand each time. The submodule option exists precisely for that, since you can bump the pinned commit when you are ready rather than chasing main.
The licence is MIT, which permits commercial use, modification and redistribution provided the copyright notice and permission notice are retained. The README defers to the LICENSE file for the exact terms. That is the whole of what the repository states on licensing, and it is not legal advice: if you are embedding these prompts in a product you ship, read LICENSE yourself and check how your fork handles attribution. Nothing in the material suggests any restriction tying the skills to the BlackTwist service, but the MCP integration is the part that depends on an external account.
Editorial conclusion
Adopt it if you already run Claude Code or Cursor and want your social writing to inherit a fixed voice and audience definition instead of re-explaining yourself in every prompt. Skip it if you need an autonomous scheduler: without the BlackTwist MCP server connected, every skill drops to advisory mode and you post by hand. Before installing, open skills/social-media-context-sms/ and confirm you are willing to maintain that file, because it is the input every other skill depends on.
Community notes