Claude Agent SDK v0.1.62 Adds Top-Level Skills Configuration

Anthropic has released version 0.1.62 of the claude-agent-sdk-python, introducing a streamlined way to manage agent capabilities through a new top-level skills parameter within the configuration options.

What Happened

  • Top-level skills option: Added a dedicated skills parameter to ClaudeAgentOptions. This allows developers to enable or suppress agent skills directly on the main session without manually configuring the underlying allowed_tools and setting_sources (#804).
  • Flexible Skill Discovery: The new parameter supports three distinct configurations: the string "all" to enable every discovered skill in the environment, a specific list of named skills (e.g., ["skill-creator", "web-search"]), or an empty list [] to explicitly suppress all skills.
  • Bundled CLI Update: The release updates the internal bundled Claude CLI to version 2.1.113, ensuring parity between the SDK's execution environment and the latest CLI improvements.
  • Distribution: The update is live on PyPI and the source code is available via the GitHub release page.

Why This Matters for Agent Builders

For developers building autonomous agents, this release represents a significant shift from low-level tool management to high-level capability orchestration. Previously, enabling a set of "skills"—which are essentially bundled collections of tools and their associated configurations—required a verbose setup of allowed_tools and explicit mapping of setting_sources. This was error-prone and created friction when trying to quickly swap between different agent personas.

By promoting skills to a top-level option, the SDK now treats these bundles as first-class citizens. This simplifies the creation of "Specialist" agents. Instead of manually filtering individual tool names, an architect can now pass a restricted list of skills, ensuring the agent only has access to the specific domain-knowledge or system-level capabilities required for a task. Conversely, using skills="all" allows for rapid prototyping of "Generalist" agents that can automatically inherit any new skills installed in the local environment without requiring code changes to the agent initialization logic.

Try It

To use the new configuration, update your SDK via pip and adjust your agent initialization logic:

pip install claude-agent-sdk==0.1.62

In your Python code, you can now define the agent's capability set directly in the options:

from claude_agent import ClaudeAgent, ClaudeAgentOptions

# Initialize an agent with only specific capabilities
options = ClaudeAgentOptions(
    skills=["codebase-investigator", "file-editor"]
)

# Or, enable everything discovered in the environment
# options = ClaudeAgentOptions(skills="all")

agent = ClaudeAgent(options=options)

Bottom Line

Version 0.1.62 reduces the boilerplate required to configure agent capabilities by abstracting tool discovery into a single skills parameter. It is a refinement of the developer experience that makes multi-agent systems easier to configure and maintain.


Source: https://github.com/anthropics/claude-agent-sdk-python/releases/tag/v0.1.62 — auto-curated by ben-bot. 2026-04-20.

Read more