← Back to Skills Marketplace
Reproduce
by
sparky0520
· GitHub ↗
· v1.0.4
· MIT-0
158
Downloads
0
Stars
0
Active Installs
5
Versions
Install in OpenClaw
/install recursive-spawn
Description
Enables an Openclaw agent to spawn sub-agents (child Openclaw instances) when a task is too large, complex, or parallel to handle alone. Use this skill whene...
Usage Guidance
This skill is coherent for its stated purpose, but take these practical precautions before installing or using it:
- Provide only a non-production API key when testing; do not include keys in payloads or snapshots. The script will use your provider key (env var or api_key override) to call model APIs.
- Sanitize progress_so_far carefully: it is sent to the provider and becomes part of the child's prompt. Remove secrets, credentials, personal data, and anything you wouldn't want sent to the model provider.
- Minimize tools passed to children. File-access tools grant children read/write access to arbitrary paths — prefer returning results in text rather than giving file tools unless necessary.
- Note the manifest declares litellm as a pip dependency but there is no automated installer. Ensure your runtime has the correct Python version and litellm installed (pin versions if you need reproducibility).
- Address the minor metadata inconsistency: the registry metadata said 'Required env vars: none' while SKILL.md/manifest say at least one provider key is required at runtime. Confirm how your deployment supplies keys (env vs per-call) before use.
If these points are acceptable and you follow least-privilege practices (sanitize snapshots, avoid providing file tools unless necessary, use test keys), the skill appears to be what it claims to be.
Capability Assessment
Purpose & Capability
The skill is explicitly a spawner for child Openclaw agents. The included Python script, use of LiteLLM, and the manifest's provider API-key entries all align with that purpose. Asking for a provider API key (Anthropic/OpenAI/Gemini/etc.) is expected for making model API calls.
Instruction Scope
Instructions are narrowly scoped to creating child agents, enforcing a payload schema, enforcing a max-depth, and warning about sanitizing snapshots and tool permissions. This is appropriate, but be aware: progress_so_far is explicitly sent to the provider and injected into the child's context (data-exposure risk). The skill also allows passing file-access tools to children (which grants read/write access to arbitrary paths) — this is documented but a significant capability to grant and should be handled with least privilege.
Install Mechanism
The skill is instruction-only (no install spec) and includes a Python script. The manifest lists a pip dependency on litellm>=1.40.0. No opaque downloads or extract/install steps are present. Note: because there's no automated install spec, the environment running the skill must already have (or must install) the litellm package and Python runtime.
Credentials
The manifest and SKILL.md correctly describe that at least one provider API key is required at runtime (or an Ollama local setup). However, the registry summary earlier showed 'Required env vars: none' which is inconsistent with the manifest/SKILL.md. The script also accepts an api_key override via litellm_kwargs, so credentials can be supplied either via env vars or per-call overrides — this is flexible but means callers should avoid passing raw keys in spawn payloads or snapshots.
Persistence & Privilege
The skill does not request always: true, does not modify other skills, and is user-invocable with normal autonomous invocation allowed. The spawn capability increases blast radius (child agents making provider calls), but this is the intended function and is constrained by MAX_DEPTH and the documented payload/schema rules.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install recursive-spawn - After installation, invoke the skill by name or use
/recursive-spawn - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.4
**Added multi-provider model support and switched to LiteLLM for spawning sub-agents.**
- Now supports Anthropic, OpenAI, Gemini, Groq, Mistral, and Ollama via the LiteLLM Python package.
- Provider and model are configurable with `model=` (default: "anthropic/claude-opus-4-6").
- API key environment variable for the provider must be set (e.g. `ANTHROPIC_API_KEY`, `OPENAI_API_KEY`); Ollama requires no key.
- Tools passed to children must be in OpenAI function-call format (LiteLLM handles translation).
- `spawn_openclaw.py` now requires `litellm` as a dependency.
- Updated security and environment documentation to reflect multi-provider and tool format changes.
v1.0.3
Version 1.0.3
- Added a manifest.json file for the skill.
- No changes to functionality or documentation in SKILL.md.
- This release improves metadata and discoverability without altering user-facing features.
v1.0.2
No user-visible changes detected in this version. No file changes were made.
v1.0.1
- Added explicit requirements for the ANTHROPIC_API_KEY environment variable (including doc and security note).
- Documented security considerations: environment variable handling, risks of tool/file access delegation, and sensitive data exposure in snapshots.
- Added guidance to sanitize progress snapshots before spawning child agents.
- Changed the SKILL.md front matter to include new "requires" and "security" fields.
- No code changes in this version; documentation-only update focused on operational security and correct environment configuration.
v1.0.0
openclaw-spawner 1.0.0
- Initial release of the Openclaw Spawner skill.
- Enables Openclaw agents to spawn child agents to handle large, complex, or parallelizable tasks.
- Provides a standard protocol and Python helpers for spawning, context passing, and result retrieval.
- Enforces a configurable spawn depth limit (default MAX_DEPTH=3).
- Includes detailed guidelines, payload schema, and spawning strategies for safe, effective task delegation.
- Sample code and step-by-step instructions for sequential and parallel spawning included.
Metadata
Frequently Asked Questions
What is Reproduce?
Enables an Openclaw agent to spawn sub-agents (child Openclaw instances) when a task is too large, complex, or parallel to handle alone. Use this skill whene... It is an AI Agent Skill for Claude Code / OpenClaw, with 158 downloads so far.
How do I install Reproduce?
Run "/install recursive-spawn" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Reproduce free?
Yes, Reproduce is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Reproduce support?
Reproduce is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Reproduce?
It is built and maintained by sparky0520 (@sparky0520); the current version is v1.0.4.
More Skills