ML.ai CodeSkills & Commands

Skills

The five bundled skills, how to attach one to a message, and how project-level skills show up alongside them.

Attaching a skill

Type / to open the command list, then /skills, or open Settings → Skills. Selecting a skill attaches a chip to your next message, and you type the actual task next to the chip. Removing the chip removes the skill from that pending message; it doesn't cancel anything already sent.

Bundled skills

SkillWhat it does
commitWrites a Conventional Commits message from the current diff
create-subagentComposes a new custom subagent from a plain-language description
ml-ai-proofExplains, in plain language, what a run's verification evidence actually establishes
project-pdfGenerates a PDF overview of the current project
site-redesign-ui-uxDesigns, audits, or implements production-ready web interfaces

Project and user-level skills

Any OpenCode skills configured at the user or project level appear in the same catalog, read live from the engine, so you don't need to register them with ML.ai separately.

A skill with malformed frontmatter is silently skipped by the engine rather than shown as broken. If a skill you expect to see is missing, check its frontmatter before assuming ML.ai dropped it. See troubleshooting.

Example

/commit

then send. This attaches the commit skill chip to your message; sending it with no other text is enough, since the skill's own instructions supply the task. ML.ai reads the current diff and proposes a Conventional Commits message for it, which you can edit before actually committing.