A robust framework for orchestrating AI subagents with strict safety invariants and persistent, multi-round context tracking.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install safe-subagent-spawn
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 safe-subagent-spawn 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 Safe Subagent Spawn?
Safe Subagent Spawn is a specialized skill designed to manage the lifecycle of AI subagents through a controlled wrapper rather than direct, unvalidated execution calls. By utilizing these Openclaw Skills, parent agents can delegate complex tasks to workers while enforcing strict invariants such as runtime constraints, mandatory cleanup modes, and validated model identifiers. This approach prevents common errors in agent orchestration and ensures that every delegation is secure and predictable.
The core value of this skill lies in its persistent context management. Instead of volatile session data, it maintains a full-text history in dedicated Markdown files. This ensures that multi-round tasks remain coherent and that the parent agent can track progress without cluttering its own immediate context window, effectively standardizing how sub-tasks are handled within the Openclaw Skills ecosystem.
Safe Subagent Spawn Use Cases
- Delegating specialized research or coding tasks to detached helper agents.
- Managing long-running, multi-step workflows that require persistent state across rounds.
- Creating a permanent, human-readable audit trail of all subagent interactions.
- Orchestrating sequential tasks where each step depends on the verbatim output of a previous subagent.
How Safe Subagent Spawn Works
- The parent agent initializes a task by creating a dedicated context file using a unique, validated task slug.
- Instructions and background information are injected into the context via localized Openclaw Skills scripts.
- The skill generates a strictly validated JSON payload for spawning, ensuring that fields like agentId and model are correctly mapped.
- The subagent is spawned and executes its directive, while the parent waits for a verbatim response or a timeout signal.
- Upon completion, the subagent's full output is appended to the context file, allowing for subsequent rounds of delegation if the task requires further iteration.
Safe Subagent Spawn Setup
To integrate this skill into your environment, ensure the provided Python scripts are accessible within your Openclaw Skills directory. No manual configuration of the spawning logic is required as the scripts enforce all safety rules.
# Create a new context for a specific task
scripts/create_context.py \
--task "security-audit-v1" \
--background "Analyze the authentication module" \
--directive "Check for potential SQL injection points"
Safe Subagent Spawn Data Schema & Taxonomy
This skill utilizes a structured file-based system for tracking subagent activity. All data is stored in the sub-agents/ directory using a standardized metadata taxonomy for Openclaw Skills.
| Section | Function |
|---|---|
| Metadata | Tracks creation time, task slug, and absolute file paths. |
| Standing Instructions | Enforces read-only rules and prevents recursive spawning by children. |
| Background | Stores the original request and necessary technical context. |
| Directive | Contains specific, round-based instructions for the subagent. |
| Child Output | Stores the full, verbatim response from the subagent for audit and reuse. |
| External Message | Allows for incremental information injection between active rounds. |
Safe Subagent Spawn Advanced Features
- Multi-round continuity: Seamlessly manage sequential agent interactions within a single logical task file.
- Safety Invariants: Hard-coded rules prevent the accidental use of dangerous runtime modes or incorrect model aliases.
- Deduplication Logic: Automated checks prevent redundant data from being appended to context files, maintaining efficiency in Openclaw Skills.
- Deferred Directives: The ability to gather external data before finalizing the instructions for a subagent.
- Permanent Audit Trail: Context files are never deleted, providing a persistent record of all delegated agentic logic.
SKILL.md
Loading
Related Openclaw Skills
Find Skills
The primary discovery and installation tool for expanding AI agent capabilities via the open skills ecosystem.
Freelancer Business Autopilot Lite
A lightweight business assistant that generates professional invoices and client update emails from plain-language descriptions.
Market Insight Agent
An intelligent market analysis agent designed to monitor global web novel trends, platform policies, and competitor data to drive data-backed content strategy.
Camping and Glamping Spot Finder
A specialized AI agent skill for discovering outdoor camping grounds and luxury glamping sites with real-time booking capabilities via the flyai CLI.
x402 Virtual Card Skill
The x402 skill enables AI agents to purchase and manage virtual Visa or Mastercard debit cards using USDT on the Binance Smart Chain.
May Vomit
An absurdly dramatic anti-flirting defense skill that triggers chaotic, comedic rejections when users attempt to romance the AI.
Featured*
