Skills
Add reusable Agent Skills, keep them private, or share them with everyone in your workspace.
Skills are reusable instruction packs that the agent can load. They live on the filesystem and can remain private to you or be shared with the entire workspace.
Endstack follows the open Agent Skills format. Refer to agentskills.io for the standard structure and authoring guidance for SKILL.md packs.
Where to find skills
Open Settings → Skills to view and share installed skills.
Skill files are stored under ~/.agents/skills. Each skill typically has its own directory containing a SKILL.md file and any related files it needs.
Controls
| Control | What it does |
|---|---|
| Refresh | Reloads the available skill list from disk and workspace resources. |
| Your skills | Lists skills you own, including each skill's name, description, directory, and ownership. |
| Shared with everyone | Toggles one of your skills between private and workspace-wide availability. |
| Shared with you | Shows a read-only list of skills that other members have shared. |
Settings displays an empty state when you do not have any skills or when no other workspace member has shared one with you.
Create or edit a skill
Skill content is managed through the filesystem rather than a form in Settings. Create or edit the files with Writer, EndCode, Files, or another filesystem tool, then select Refresh in Settings → Skills.
Follow the Agent Skills standard when structuring the skill's SKILL.md and supporting files.
Privacy and workspace sharing
- A private skill remains available on your instance until you share it.
- Turning on Shared with everyone makes a skill you own available to agent runs across the workspace according to workspace sharing rules.
- Skills under Shared with you are owned and shared by another member, so the list is read-only.
How skills affect the agent
An available skill gives the agent reusable instructions and workflows that it can load during a run. Skills complement built-in tools and tools supplied through MCP; sharing a skill does not change the autonomy policy that governs tool calls.