Overview
Skills package reusable instructions, scripts, and resources that agents can use for specialized tasks. Use a skill when you want an agent to follow a repeatable workflow, apply team-specific conventions, or use supporting files without repeating the same guidance in every prompt. Poolside supports the Agent Skills format. Each skill is a folder with aSKILL.md file that describes when to use the skill and how the agent should apply it. A skill can also include supporting files, such as scripts, references, and templates.
When to use a skill
Use a skill when you want agents to:- Follow a repeatable workflow, such as reviewing code against team conventions
- Apply specialized domain guidance for a task
- Use supporting files such as templates, reference docs, or helper scripts
- Reuse the same procedure across projects or team members
- Broad agent behavior that should apply to every task. Use AGENTS.md instructions instead.
- Project-specific coding rules that should live with the project. Use
AGENTS.mdinstead. - External tools or services that agents need to invoke. Use an MCP server instead.
How skills work
Skills can apply automatically when the description inSKILL.md matches your request. You can also explicitly select a skill from the skills menu for the current prompt.
The description in SKILL.md determines when the agent applies the skill. When the agent uses a skill, it reads the skill instructions and any supporting files it needs.
Local skill sources
Poolside scans the following local directories for skills:~/.config/poolside/skills/.poolside/skills/.agents/skills/
| Source | What it’s for |
|---|---|
| Project Skills | Skills stored with the current project |
| Global Skills | Personal skills available across projects |
Skill format
A skill must be a directory with aSKILL.md file. The directory name must match the name field in the SKILL.md frontmatter. The SKILL.md file must include YAML frontmatter with a name and description, followed by Markdown instructions.
description helps the agent decide when to use the skill. Write it so it describes both what the skill does and when the agent should use it.
Do not rely on the Agent Skills allowed-tools or compatibility fields to restrict tools or where a skill can run. Poolside does not enforce those fields.
For the full skill format, see the Agent Skills specification.
Skills and sandboxes
If you use local sandboxes, prefer project-level skills in.poolside/skills/ when a skill must be available inside the sandbox. User-level skills in ~/.config/poolside/skills/ are not mounted in local sandboxes. Only workspace directories and user-configured extra mounts are available inside the container.
Use a skill
Use the skills menu when you want to see available skills or explicitly apply one to your current prompt. Prerequisites- You have an active interactive
poolsession. - The skill is available from a local skill directory.
- In an interactive
poolsession, type/skills. - Review the available skills in the menu. The menu groups skills by source, such as Project Skills and Global Skills.
- Select the skill you want to use.
- Finish your prompt and send it.
Create a local skill
Add a local skill when you want to test a repeatable workflow in a project. Prerequisites- You have write access to the project where you want to add the skill.
-
In your project, create a skill directory under
.poolside/skills/: -
Add a
SKILL.mdfile: -
Add the required frontmatter and instructions:
-
Start an interactive
poolsession in the project. -
Ask for a task that matches the skill description, such as
Review my current changes against the code review skill.
SKILL.md does not match the expected format, verify your SKILL.md frontmatter and directory name.
Troubleshooting
If a skill did not run when you expected, check the following:- The skill is in one of the local skill directories Poolside scans.
- The skill’s
descriptionmatches the task you asked the agent to complete. - The
SKILL.mdfrontmatter is valid and thenamematches the skill directory. - The skill is available inside your sandbox, if you run the agent in a sandbox.