Social Media Skills for AI Agents

Why it's ranked here

It earns a look for structural discipline that is rare in prompt-only tooling: a shell script enforces the packaging spec before anything merges, and a bundled test suite proves the installer preserves a person's edited copies and symlinks across repeated runs, then confirms a real agent session actually discovers all seventeen skills afterward. Seventeen narrowly scoped tasks share one profile instead of each guessing at tone on its own. MIT licensed, and skills consistently state when data is missing rather than inventing a score or a fact to fill the gap.

What's good

Dependency checking is real rather than cosmetic: skills that need a background and tone profile check for it first and stop with a redirect rather than guessing at someone's voice. The scoring and research skills separate measured history from editorial judgment and label which one produced a given number, so a score never disguises a guess as data. The thumbnail and carousel skills say plainly when they can only hand back a prompt, instead of claiming to have generated a finished image.

Tradeoffs

Coverage stops at LinkedIn, Instagram Reels, and YouTube thumbnails, so a team working other networks gets nothing here. The scoring and Reel-analysis skills only move past editorial guesswork once a paid scraping service and a separate image account are connected, so a meaningful slice of the advertised capability sits behind credentials and cost the user has to supply themselves. Several skills are also written to launch straight into a multi-question interview the instant they trigger, skipping any explanation of what is about to happen, trading a beat of transparency for speed.

How to use it well

Run the background and tone interview first, since almost every other skill checks for its output and refuses to continue without it. From there this suits a solo LinkedIn creator or small team running a full pipeline, from drafting through analytics, more than someone who wants one throwaway post. Connect the scraping and image services only once the scoring and visual skills need to move past editorial estimates and bare prompts. It is not a substitute for a team covering other platforms or doing video editing.

Observed

License

MIT licensed

Packaging

Distributed as 17 uniformly structured skill folders, each with an instruction file carrying YAML frontmatter, and optional reference, script, and asset subfolders, per the Agent Skills spec

Structural

No application code beyond one bash validation script and one Python test script; the skills themselves are markdown instruction files, not executable programs

Packaging

Three documented install routes: a Claude Code plugin marketplace listing, individual zipped skill upload for Claude Desktop, and a copied or symlinked project folder for Codex

Structural

Drafting and formatting skills require no external API; the post-scoring and Reel-analysis skills call an optional third-party scraping service and an optional image-generation API

Structural

Bundled test suite installs into an isolated HOME directory and checks that a customized file, an unrelated folder, and a symlink all survive a second install run without being overwritten