What are skills

The MCP connector gives an assistant the tools. A skill teaches it how to use them: the sequencing, the terminology, and the safety rails that separate a useful campaign from a pile of API calls.

Twenty-two skills are available in celeryhq/simplified-ai, for Claude, Codex, ChatGPT, Cursor, and compatible agents, in two groups:

How they compose

The workflows chain together, and operators do the work underneath:

plan or repurpose content
  → generate reusable assets
  → create channel-native drafts
  → review
  → confirm
  → schedule or queue
  → analyze results

For a recording-to-social workflow, repurpose-video produces rendered clips, content-repurposer prepares grounded channel copy, and simplified-social saves the requested drafts. Use transcribe-media when a standalone transcript or subtitles are needed; generate-voiceover creates separate scripted narration.

Using one

With the plugin installed, just ask for the outcome:

Repurpose this blog post into a LinkedIn post, an Instagram carousel and three tweets, in our brand voice.

Where the client loads installed skills, the assistant follows content-repurposer, reads accessible source and brand context, and prepares the requested drafts. Image generation happens when requested or authorized; a connector alone does not install this repository’s skills.

The draft-before-publish boundary

Social drafting stays separate from publication. Publishing requires explicit authorization for the content, accounts, timing, and comments; an unchanged approved plan does not need repeated approval. For a draft request, cross-platform-campaign prepares the rollout without publishing. For an explicitly authorized unchanged publication plan, it proceeds without asking for the same approval again. That boundary is the reason these are skills rather than raw tool access.

Skills and the API

Everything a skill does is available directly through the API. The skill encodes the sequencing so you don’t have to.