1os

1os ships the agent harness by 1ovate: the 1ovate doctrine, the skills that answer to it, and shared output styles.

Skills

Nineteen, each one file.

House

• doctrine — the house rules every other skill answers to. Read at the start of a session, and whenever you are unsure what to do.
• brief — you are agreeing what work is before anyone builds it. A BRIEF.md committed beside the code, instead of an issue tracker: the gap that code closes, and the features that count as done.

Missions

• mission — work arrives, is under way or has come back. Picks the skills to run and who runs them, marks how far an artefact has got against its own Done when, and decides how hard to verify what comes back.
• avatar — a charted skill is running. Holds the agent to its chart's states, moves and tools by hooks, halts it to escalate, resumes it on your command, and binds the agents it starts.
• spawn — you are handing work to an agent rather than doing it. Scopes the instruction, and picks who runs it.
• checkpoint — you are out of context and a fresh session must pick the work up. Writes what the brief and the commits cannot carry to a temp dir, and finishes nothing on the way.
• loadout — you are starting a repository, or picked one up cold. What it must state about itself, the brief beside its code, the commands it runs from, and the container it ships as.

Library

• cynefin — work arrives and nobody has said what practice it needs. Triages it by Cynefin domain: what the domain needs, and where to stop. Writes no files.
• markdown — you are writing or cutting any document, or a script must read one by its structure. Only the lines a reader acts on, what renders everywhere, what a table is for.
• gherkin — you need one capability specified before anyone builds it. A .feature.md beside the code: the capability, its scenarios, how each step is checked, and the stages from idea to passing steps.
• mermaid — you are checking a skill's chart. Its shape in Mermaid beside its contract, and the lint that reports every problem with its fix.
• marp — you are turning a document into slides. The argument, the slide, the Marp rendering with the brand template.
• git — you are giving an agent write access, or wiring CI. Trunk protection, the agent's own identity, split credentials, build and release workflows, private versioned images, dependency updates, and several agents on one tree.
• kev — you have many short texts to sort or label, or a hook or script could check them. Fixed questions asked of small local models, probed before they are trusted, no tokens spent.
• deploy — you are putting a version on the web, or taking it off. The stacks repository Portainer watches, shared Traefik routing, secrets outside git, staging before production, and rollback by revert.

Lean

• kata — you need a skill specified. The SKILL.md itself: the capability and its scenarios, and its chart.
• dojo — you need to know whether a skill works. How each of its steps is checked, and what its runs say.
• kaizen — a skill misled an agent, or you want to know what in 1os needs fixing. A report filed where it happened, a harvest of the sessions for the rest, and each finding proposed as the smallest fix to the line that governed the agent.
• sensei — you want a plan stress-tested before you build it. Interviews you one question at a time until the design holds.

Output styles

Pick one with /output-style; they are listed under the plugin name. Both restyle Claude Code and leave its coding behaviour alone.

• 1os:ELI5 — plain language, short sentences, no jargon. For when your brain is fried.
• 1os:essentials — a peer, not an assistant. Symbol-led finding blocks (👁️ observed, ✅ changed, 🧪 evidence, 👉 your turn) instead of prose, so the work is fast to review and accept.

Doctrine

skills/doctrine/SKILL.md governs every skill, itself included, and no skill may edit it. It is a skill so that it routes and so that it can be graded; a change to it puts every skill in the eval blast radius.

Contributing

AGENTS.md is the whole procedure — and an agent working in this checkout has already loaded it, so you can just ask.