[DOCS] [SDK] Missing "Skill" tool documentation in Tool Input/Output Types reference

Resolved 💬 3 comments Opened Jan 18, 2026 by coygeek Closed Feb 28, 2026

Documentation Type

Missing documentation (feature not documented)

Documentation Location

  • docs/en/agent-sdk/python.md - docs/en/agent-sdk/typescript.md - Reference link: docs/en/agent-sdk/skills.md

Section/Topic

The docs/en/agent-sdk/skills.md file states: > "Enable Skills by including "Skill" in your allowed_tools configuration."

Current Documentation

The docs/en/agent-sdk/skills.md file states:

"Enable Skills by including "Skill" in your allowed_tools configuration."

However, in the comprehensive tool reference sections of python.md and typescript.md, the list of built-in tools includes:

  • Task
  • AskUserQuestion
  • Bash
  • Edit
  • Read
  • ... (and several others)

The tool Skill is entirely omitted from these reference lists.

What's Wrong or Missing?

While the documentation explains how to enable the Skill tool, it fails to provide the technical schema (Input/Output format) that developers encounter when Claude actually invokes a Skill.

Since the SDK requires developers to iterate through AssistantMessage content blocks (which may contain tool_use for the Skill tool), the lack of a documented schema makes it difficult to handle these blocks programmatically or to understand what parameters Claude is passing to the Skill dispatcher.

Suggested Improvement

Add the Skill tool to the "Tool Input/Output Types" section in both SDK references. Based on the "Agent Skills" architecture, the entry should look similar to this:

Skill

Tool name: Skill

Input:

{
    "skill_name": str,       # The unique identifier/name of the skill being invoked
    "rationale": str         # Claude's reasoning for why this skill is being used
}

Output:

{
    "output": str,           # The result or message returned after the skill execution
    "is_error": bool         # Whether the skill execution encountered a failure
}

(Note: Please verify the exact field names, such as skill_name vs name, against the latest CLI control protocol.)

Impact

High - Prevents users from using a feature

Additional Context

  • Related Documentation: In docs/en/agent-sdk/skills.md, there is a section titled "Tool Restrictions" that explicitly mentions using the allowedTools option for the Skill tool, but provides no link to a schema.
  • Consistency: The Task tool (which handles subagents) is well-documented in the reference. Since Skill is the other primary way Claude delegates complex or filesystem-based logic in the SDK, it deserves an equivalent reference entry.

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗