Skip to main content
Skills are reusable behavior modules (SKILL.md) that let agents execute complex workflows consistently.

​
Why skills matter

  • Convert repeated prompts into stable operating procedures
  • Keep behavior portable across sessions and channels
  • Enable autonomous agents to self-document successful workflows

​
Skill storage and precedence

Runtime resolves skills by layered priority:
  1. Workspace skills
  2. Agent-created self skills
  3. User-global skills
  4. Builtin skills
Agent-created skills are stored in: 
/root/.ara/workspace/self-skills/<skill-name>/SKILL.md

​
Skill lifecycle

1

Discover

Agent searches registries (for example ClawHub) or local skill paths.
2

Install or create

Skill is installed or authored through skill tools and written as SKILL.md.
3

Load into context

Runtime includes skill summaries and reads full files when needed.
4

Execute and refine

Agent executes with real tasks, then can update skill content for better future performance.

​
Auto-create skills pattern

A strong pattern for autonomy is:
  1. Agent completes a multi-step task manually
  2. Agent saves the successful pattern as a new skill
  3. Future tasks call that skill directly
  4. Agent revises the skill when edge cases appear
This is the bridge from one-off execution to reusable intelligence.

​
Good skill design

  • Keep scope narrow and explicit
  • Include inputs, outputs, and constraints
  • Add failure handling and fallback paths
  • Prefer composable skill chains over giant monolithic skills

Learning loop

See how memory and reflection improve skills over time.

Multi-agent teams

Delegate skills across specialized subagents.