Load context on-demand instead of cramming everything into the system prompt.
What Are Skills?
Skills are reusable packages of instructions and context that agents retrieve when needed. Instead of a massive system prompt, your agent sees a list of available skills and a skill_get tool to load the full content when relevant.
Quick Start
From the Registry
- Open your agent's settings → Prompt tab
- Scroll to Skills
- Click Browse → search and select a skill
- Done — the skill's name, description, and URL are filled in automatically
By URL
Click + URL and provide:
| Field | Value |
|---|---|
| Name | Identifier the agent uses (e.g. api-docs) |
| Description | When to load this skill — the agent sees this |
| URL | https://api.inference.sh/skills/{ns}/{name}/content or any HTTPS URL |
Inline
Click + Text and write the content directly. Good for short, agent-specific instructions.
Preloading
By default, skills are loaded on-demand via skill_get. For skills the agent needs on every conversation, enable preload — the content is injected directly into the system prompt at startup.
Toggle Preload on the skill in the Prompt tab, or set it in YAML:
1skills:2 - name: pricing-rules3 skill_id: myteam/pricing-rules4 preload: truePreloaded skills are available from turn one — no tool call needed. Use this for core reference material the agent always needs. Keep large or situational content as on-demand.
How It Works at Runtime
On-demand skills appear in the skill_get tool listing. The agent calls skill_get when the conversation matches a skill's description:
1skill_get(skill="code-review") → main instructions2skill_get(skill="code-review", path="references/checklist.md") → supporting filePreloaded skills are already in the system prompt — no tool call needed.
Content is cached per session — loading the same skill twice doesn't re-fetch.
Learn More
- Creating skills — build and publish your own
- The skill registry — browse and share public skills
- Using with other agents — install to Claude Code, Cursor, and more
- Skills overview — how skills work, the standard, and inference.sh's approach