Tools tell an agent what it can do. Skills tell it how.

What a skill is

A short text: when to use it, which tools to call in which order, what to check with the user. The format is the one the open agent-skills ecosystem uses, a SKILL.md with a name and a one-line description. The agent sees the name and description in its tool list and reads the full text only when it decides the skill applies.

How Clemali serves it

Every skill attached to an environment becomes a tool named skill__<slug>. Calling it returns the instructions. That is the same progressive disclosure a skill installed on a laptop gets, in every MCP client, with nothing on the agent's machine. The environment's instructions also tell the agent that procedures exist and to read one before a matching task.

Loading a skill is free: it is not a call to one of your systems and is not counted against your plan.

Three sources

• Connector guides: one per connector — PostgreSQL, SSH, Kubernetes, Redis, S3 and the rest — written by us and served automatically with the connector. Add a PostgreSQL server and the agent gets skill__postgres, which teaches it to inspect the schema first, keep results small, and confirm before a write. They take no slot; one switch per environment turns them off.
• Public skills: SKILL.md files found on GitHub, sorted by stars. Each is pinned to the exact file it was read at; when the original changes, the new text waits for our review before it is served. The ones we have read carry a verified mark.
• Your own: written in the console, private to your organization, attachable to any of your environments. This is where your conventions live: which database is the source of truth, what a valid invoice looks like, who to ask before a refund.

Attach one

Open the environment, tab Skills, Add skill. Pick from the catalogue or write your own. Agents connected to the environment see it at their next session. Disable a skill to keep it attached but hidden; remove it to take it out.

Free allows five skills per assistant; Team and Enterprise have no limit. Plans →

Install on a machine instead

If you prefer skills on the agent's machine, the same content is on GitHub for the skills CLI:

npx skills add thomasdseao/exorails-skills
npx skills add thomasdseao/exorails-skills --skill postgres

And an environment's enabled skills, including your own, download as a folder you install with npx skills add ./exorails-<env>-skills. The download needs your session; no public address ever describes an environment.

Writing a good one

• The description decides whether the skill is loaded. Say when it applies, in one line. The body says how.
• Name the tools by the names the agent sees, with their server prefix: prod-db__query, not “the database”.
• Put the reads before the writes, and say what to confirm with the user before anything irreversible.
• Keep it under a few kilobytes. A skill is read in one go; a manual is not.

Open an environment →