An end-to-end orchestration skill for managing Anthropic Claude Managed Agents through a robust Python helper or the native ant CLI.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install claude-managed-agents
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install claude-managed-agents using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Claude Managed Agents?
The Claude Managed Agents skill provides a high-level interface for interacting with Anthropic's managed-agents beta. It simplifies the complexity of the Files API, session management, and environment configuration by offering an SDK-first Python helper. This ensures developers can safely create, update, and monitor agents without manually constructing raw REST requests.
By leveraging Openclaw Skills, users gain access to a standardized workflow for handling long-running agentic tasks. The skill manages the intricate lifecycle of agent versions, secure environments with limited networking, and real-time event streaming for steering agent behavior mid-run.
Claude Managed Agents Use Cases
- Creating persistent AI coding assistants with access to specific Python packages and system instructions.
- Provisioning secure, sandboxed execution environments with restricted network access to specific hosts.
- Managing complex session states including file uploads, resource mounting, and artifact downloads.
- Real-time steering of agent behavior through interruptions and manual tool confirmations.
- Automated cleanup and archiving of agent resources to maintain environment hygiene.
How Claude Managed Agents Works
- Preflight Check: The skill uses a
doctorcommand to verify API keys and connectivity to Anthropic's managed-agents beta endpoint. - Resource Provisioning: Users define the Agent (intelligence) and the Environment (compute/capabilities) via the Python helper CLI.
- Session Initiation: A session is created, linking an Agent to an Environment, often including mounted file resources from the Files API.
- Interaction Loop: User messages are sent to the session, triggering the agent's logic and tool-use capabilities.
- Monitoring & Steering: The system streams live SSE events, allowing for manual interrupts, tool approvals, or providing custom tool results.
- Lifecycle Management: Upon completion, the skill facilitates archiving or hard deletion of sessions and environments.
Claude Managed Agents Setup
Ensure your environment is configured with the necessary API credentials before using Openclaw Skills:
# Set your Anthropic API Key
export ANTHROPIC_API_KEY='your-api-key-here'
# (Optional) Set the beta header if not using the default
export ANTHROPIC_MANAGED_AGENTS_BETA='managed-agents-2026-04-01'
# Run the preflight doctor to verify setup
python3 ~/.openclaw/skills/claude-managed-agents/scripts/managed_agents.py doctor --live
Claude Managed Agents Data Schema & Taxonomy
The skill organizes its operations around several core entities and metadata structures:
| Entity | Primary Identifiers | Description |
|---|---|---|
| Agent | agent_id |
Defines the model, system prompt, and built-in toolsets. |
| Environment | env_id |
Defines networking rules, pip packages, and runtime constraints. |
| Session | sess_id |
An active instance of an agent running in an environment. |
| Files | file_id |
Binary or text data uploaded to the Files API for mounting. |
| Events | event_id |
Granular logs of agent thoughts, tool calls, and user interactions. |
Claude Managed Agents Advanced Features
- Dynamic Resource Mounting: Add or remove files and resources from an active session without restarting the agent.
- Multi-Backend Support: Toggle between
sdkmode for native performance andhttpmode for environment compatibility. - Steering Controls: Support for
user.interruptto redirect agents mid-process anduser.tool_confirmationfor human-in-the-loop workflows. - Network Guardrails: Explicit host white-listing (e.g.,
api.example.com) to prevent data exfiltration in production environments. - Session Artifact Recovery: Automated workflows for downloading generated files and logs from completed agent runs.
SKILL.md
Loading
Related Openclaw Skills
Backup Rotator
A robust Python-based tool for managing automated backup rotation, retention policies, and file integrity verification.
API Tester
A lightweight Python-based toolkit for testing, validating, and benchmarking REST API endpoints without external dependencies.
System Doctor (sys-doctor)
A comprehensive system diagnostics and health check tool for Linux servers that generates detailed reports in terminal, HTML, or JSON formats.
BrickLink AFOL Skill
A professional-grade CLI integration for the BrickLink API that enables automated Lego inventory management, order processing, and market data retrieval.
Gen Z Lingo Coach
A comprehensive linguistic tool for decoding, translating, and auditing Gen Z slang and internet culture phrasing with professional restraint.
Running Tracker
An automated fitness tracking skill that logs running data, calculates performance metrics, and provides progress insights via a localized markdown database.
Featured*
