> Fetch clean Markdown by appending `.md` to any page URL under https://signalwire.com/docs or requesting it with the HTTP header `Accept: text/markdown`. The root index at https://signalwire.com/docs/llms.txt lists the available documentation indexes.

# ClaudeSkillsSkill

> Load Claude Code-style SKILL.md files as agent tools.

[add-skill]: /docs/server-sdks/reference/typescript/agents/agent-base/add-skill

Load Claude Code-style SKILL.md files as agent tools. Each SKILL.md file in the
configured directory becomes a SWAIG function, with YAML frontmatter parsed for
metadata.

**Class:** `ClaudeSkillsSkill`

**Tools:** Dynamically created from SKILL.md files (prefixed with `tool_prefix`)

**Env vars:** None

**Multi-instance:** Yes

**`skills_path`** `string` — required

Path to the directory containing Claude skill folders (each with a SKILL.md file).

---

**`include`** `string[]` — default: \*

Glob patterns for skills to include.

---

**`exclude`** `string[]`

Glob patterns for skills to exclude.

---

**`tool_prefix`** `string` — default: claude\_

Prefix for generated tool names. Use an empty string for no prefix.

---

**`prompt_title`** `string` — default: Claude Skills

Title for the prompt section listing skills.

---

**`prompt_intro`** `string` — default: You have access to specialized skills. Call the appropriate tool when the user's question matches:

Introductory text for the prompt section.

---

**`skill_descriptions`** `object`

Override descriptions for specific skills. Keys are skill names, values are
description strings.

---

**`response_prefix`** `string`

Text to prepend to skill results.

---

**`response_postfix`** `string`

Text to append to skill results.

---

**`allow_shell_injection`** `boolean` — default: false

Enable shell command preprocessing in skill bodies.
**Security warning:** allows arbitrary shell execution.

---

**`allow_script_execution`** `boolean` — default: false

Discover and list `scripts/` and `assets/` files in prompt sections.

---

**`ignore_invocation_control`** `boolean` — default: false

Override `disable-model-invocation` and `user-invocable` flags from frontmatter.
When `true`, all discovered skills are registered regardless of invocation control.

---

**`shell_timeout`** `integer` — default: 30

Timeout in seconds for shell injection commands.

---

```typescript {6-11}
import { AgentBase, ClaudeSkillsSkill } from '@signalwire/sdk';

const agent = new AgentBase({ name: 'assistant', route: '/assistant' });
agent.setPromptText('You are a helpful assistant.');

await agent.addSkill(new ClaudeSkillsSkill({
  skills_path: '/path/to/skills/directory',
  tool_prefix: 'skill_',
  include: ['*'],
  exclude: ['deprecated-*'],
}));

agent.run();
```

> Load Claude Code-style SKILL.md files as agent tools.