subagent-isolation-guard

What This Skill Does

The subagent-isolation-guard is a critical architectural utility for OpenClaw users running complex, multi-layered agentic workflows. Its primary purpose is to enforce strict boundaries between different sub-agents, preventing the common pitfalls of context leakage and unintended semantic routing. By mandating isolated workspace directories and providing a robust bypass mechanism for the routing logic, this skill ensures that a secondary or specialized agent—such as a project manager or system architect—does not inherit or corrupt the context of the parent agent.

Installation

To integrate this guardrail into your OpenClaw environment, execute the following command in your terminal:

clawhub install openclaw/skills/skills/halfmoon82/subagent-isolation-guard

Ensure that you update your AGENTS.md file to reflect the specific agentDir paths for any new or existing subagents created thereafter.

Use Cases

This skill is essential when managing hierarchical agent teams. For instance, when a 'Developer' agent is invoked by an 'Architect' agent, the isolation guard prevents the Developer from accidentally accessing the Architect's temporary scratchpad or writing files to the global root. It is also vital for preventing the semantic router from misinterpreting a subagent's status messages as instructions to switch the primary model or reset the conversation thread.

Example Prompts

1. "Initialize a new subagent named 'CodeReviewer' with isolation protocols enabled and map its workspace to agents/review/workspace."
2. "Configure the current session identifier to include the :subagent: flag to bypass global routing middleware."
3. "Verify the workspace boundaries for the active subagent to ensure no file path collisions with the main workspace."

Tips & Limitations

Always ensure that the :subagent: identifier is appended to your session keys consistently; failing to do so will result in the subagent being treated as part of the main agent thread, potentially causing the very model-switching issues you are trying to avoid. While this skill provides robust logical isolation, it assumes the user correctly configures the agentDir in their local configuration files. It does not automate the migration of existing legacy files, so move them to dedicated directories manually after installation.