A skill is a markdown file that teaches an agent a procedure: how to design something before building it, how to research a company, how to verify a page with a real browser, how to document an incident.
They are protocol-level, so they work on any brain with any model. Install one, and the agent knows a way of working it did not know before.
# as a Claude Code plugin marketplace /plugin marketplace add giobi/claude-skills /plugin install design # or with the /brain package manager, inside any brain /brain install design # or by hand — one folder, no build step curl -sL https://brainprotocol.it/skills/design/design.zip -o /tmp/design.zip unzip -o /tmp/design.zip -d .claude/skills/
Using something other than Claude Code? Hit copy prompt on any skill below and paste it into your agent. It points at a page that tells the agent how to fetch the skill and what to do with it — including what changes when there is no brain around it.
Builds shared understanding before anyone writes anything. It grills you one question at a time until the concept holds in three sentences, captures the vocabulary as it surfaces — the words you and the agent are using differently without noticing — and settles the system into a few deep modules with narrow interfaces. The output is a design.md a future session can read cold.
/plugin install design
Implements those modules under two constraints at once: TDD for the rhythm (one test, one implementation, no code without a test that demands it) and deep modules for the structure — you design the interface, the agent may write the implementation. After each area it asks whether the module can be delegated now, then loops back to update design.md so the map stays alive.
/plugin install build
Build mode — TDD + deep modules + a return loop that keeps the design alive
/plugin install build
Collaborative design — builds shared understanding of any project before acting
/plugin install design
Run one session as the conductor of the other live Claude sessions in tmux: register them, delegate with the right provider, relay decisions, check RAM before every open, and know when to send the user to the dedicated session instead
/plugin install orchestrator
Mid-session checkpoint (log, database, commit, push) without closing
/plugin install save
CRUD for private/public mini-sites and reports — auth-gated internal links, share tokens, slideshows, brand fill
/plugin install artifact
PressLess — AI static site generator. WordPress without the weight.
/plugin install pressless
DEPRECATED — superseded by the `artifact` plugin, which is where development continues. Kept installable for brains still pointing here.
/plugin install public
SharePoint access via Microsoft Graph — tenant-wide search, browse sites, read and upload files. Reuses the email skill's O365 token.
/plugin install sharepoint
WordPress management — REST API, WP-CLI, multi-site, Puppeteer, GDPR
/plugin install wordpress
Ghostwriter — write in the user's voice, not as an LLM
/plugin install ghostwriter
Gmail orchestrator — triage inbox, search, draft replies in-thread
/plugin install gmail
AI image generation — Gemini Imagen, fal.ai (image/video/audio/TTS), Replicate Flux
/plugin install imagen
Cloudflare DNS — zones, records, cache purge, R2, Pages
/plugin install cloudflare
Internet.bs — domain management: check, register, renew, DNS, nameservers
/plugin install internetbs
Learn — genera cheatsheet interattive per imparare comandi e tool tecnici
/plugin install learn
GitHub — repos, issues, PRs, gists, Pages, branches via API
/plugin install github
A skill is a folder with a SKILL.md. The frontmatter is the whole contract:
a name, a one-line description the agent matches on, and whether a human can invoke it.
Everything after it is plain markdown — when to use it, when not to, the workflow,
the fallbacks when a companion skill is missing.
Write it in English, including the name: a skill name is an address, and addresses do not get translated. Keep the "when not to use this" section honest — it is what stops an agent reaching for the wrong tool.