A robust security and stability layer for Openclaw Skills that enforces physical workspace isolation and bypasses semantic routing loops for sub-agents.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install subagent-isolation-guard
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 subagent-isolation-guard 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 Subagent Isolation Guard?
Subagent Isolation Guard is a specialized framework designed to harden the operation of complex multi-agent systems. It addresses two critical failure points in production agent environments: context contamination and recursive routing loops. By ensuring that every sub-agent operates within its own dedicated physical directory, this tool prevents file write conflicts and data leakage between different agent roles.
Furthermore, it implements a sophisticated semantic routing bypass. This allows Openclaw Skills to identify sub-agent sessions and prevent the primary agent's global routing logic from triggering unnecessary model resets or context clearing, which often happens when sub-tasks are mistaken for new user intents. This isolation ensures that each agent in the hierarchy maintains its own logical flow and state.
Subagent Isolation Guard Use Cases
- Deploying multi-role agent teams (e.g., PM, Architect, and Developer) where each requires a distinct, non-overlapping workspace.
- Preventing global semantic webhooks from interfering with the local execution logic of specialized sub-agents.
- Scaling Openclaw Skills in high-concurrency environments where multiple agents might otherwise trigger file-system collisions.
- Eliminating infinite loops or model switching caused by recursive agent calls.
How Subagent Isolation Guard Works
- The system detects sub-agent activity by scanning for the specific
:subagent:identifier within the session key. - Upon identification, the guard enforces the use of a unique, pre-configured agentDir for the sub-agent's physical operations.
- When a request hits the semantic-webhook-server, the system checks for the isolation flag.
- If the sub-agent flag is present, the server executes a Routing Bypass, returning a continue status to the orchestrator.
- This bypass skips standard injection and model suggestions, allowing the sub-agent to complete its task without external interference.
Subagent Isolation Guard Setup
To integrate this guard into your Openclaw Skills, you must configure both your agent definitions and your webhook server:
- Define unique directories for each agent in your configuration file:
mkdir -p agents/pm/workspace/
mkdir -p agents/architect/workspace/
- Ensure your sub-agent session IDs are generated with the mandatory identifier:
# Example Session ID format
SESSION_ID="task-123:subagent:coder"
- Update your semantic-webhook-server.py logic to recognize the session_key and return the bypass signal.
Subagent Isolation Guard Data Schema & Taxonomy
The Subagent Isolation Guard organizes metadata and workspace paths to ensure strict boundaries for Openclaw Skills:
| Attribute | Type | Description |
|---|---|---|
agentDir |
String (Path) | The isolated physical directory where the sub-agent reads and writes files. |
session_key |
String | A unique ID containing the :subagent: tag used to trigger routing bypass. |
allowAgents |
Array | A whitelist of authorized sub-agents allowed to operate under the isolation guard. |
bypass_action |
Enum | The fixed response (e.g., continue) sent to the webhook server to maintain session state. |
Subagent Isolation Guard Advanced Features
- Physical Workspace Partitioning: Automatically maps sub-agents to non-overlapping directory structures to prevent data corruption.
- Semantic Bypass Logic: Specifically ignores global routing triggers for sessions marked as sub-agents.
- Context Preservation: Prevents the master agent from force-resetting the context of specialized Openclaw Skills during recursive calls.
- Role-Based Guardrails: Enforces the allowAgents whitelist to ensure only approved sub-agents can execute within the isolated environment.
SKILL.md
Loading
Related Openclaw Skills
Skill Quick Index
A high-speed discovery and indexing tool that maps natural language intents to specific local AI agent functions.
Skill Priority Setup
Automated skill tiering and injection policy configuration tool for optimizing Openclaw Skills performance.
Multi-Agent Dev Team v2.2
A powerful orchestration skill for managing flexible, multi-role AI agent development teams ranging from 2 to 10 members.
Model Failover Doctor
A diagnostic and repair utility designed to resolve persistent All models failed errors and routing mismatches within the OpenClaw ecosystem.
Hitchhiker's Guide Skill
A text adventure game engine that transforms your AI agent into a witty, absurdist Game Master based on the 1984 Infocom classic and Douglas Adams' universe.
PARA Second Brain
A comprehensive organizational framework that enables semantic search across an agent's entire PARA-structured knowledge base.
Featured*
