Skills

Skills are reusable capability definitions the agent writes, reads, and refines. They are plain files — inspectable, editable, versionable, shareable. If the agent learns to do something well once, you should never have to teach it again.

Lifecycle

• Creation — after complex tasks, the agent can autonomously create a skill capturing what worked
• Improvement — skills refine themselves during subsequent use
• Invocation — browse with /skills, invoke directly with /<skill-name>
• Import — user-created skills can migrate in from other agents (see Migration)

What a skill looks like

A skill is a directory with a Markdown definition (and optional support files). Because it is a file, you can review what the agent wrote before it runs again — or improve the wording yourself and the agent picks that up.

~/.synapse/skills/
├── deploy-checklist/
│   └── SKILL.md        # the agent wrote this after a deploy task
├── release-notes/
│   └── SKILL.md        # improved 3× since creation
└── pdf-extraction/
    └── SKILL.md

Open standard

Synapse is compatible with the agentskills.io open standard, and the repository bundles a large catalog of ready-made skills (Apple suite, creative tools, GitHub workflows, MLOps, note-taking, and more) plus an optional-skills catalog.

Managing skills

# inside a session:
/skills            # browse installed skills
deploy-checklist   # invoke a skill by name

User-created skills live under $SYNAPSE_HOME/skills/. Imported skills land in ~/.synapse/skills/openclaw-imports/. Delete a directory to remove a skill — no registry to clean up.