← Back to Skills Marketplace
Recursive Swarm
by
Pedro Gonzalez
· GitHub ↗
· v0.1.1
277
Downloads
0
Stars
2
Active Installs
2
Versions
Install in OpenClaw
/install recursive-swarm
Description
Bounded recursive orchestration for complex tasks that are too large for one agent turn but cleanly decompose into a few independent subproblems. Use for mul...
Usage Guidance
This skill appears to do exactly what it says: local, bounded orchestration stored in a run folder. Before using it: (1) run the scripts in a dedicated directory (do not point runs-dir at system or sensitive folders); (2) review the Python files (they only read/write tree.json, events.jsonl, and node files) to confirm they meet your policy; (3) be cautious when creating 'ops' or 'browser' nodes or configuring executors that allow subagents/direct tools — those child executors (not these scripts) could perform network or shell actions if the agent has such tools enabled; (4) no secrets are requested by this skill, so do not supply credentials into the run directory or node artifacts. If you need higher assurance, test with a small harmless run first and inspect events.jsonl and nodes/* after execution.
Capability Analysis
Type: OpenClaw Skill
Name: recursive-swarm
Version: 0.1.1
The recursive-swarm skill bundle provides a structured framework for an AI agent to decompose complex tasks into a managed tree of sub-tasks. The included Python scripts (e.g., init_run.py, upsert_node.py, mark_node.py) are standard utilities for managing task state, audit logs (events.jsonl), and result aggregation within a local directory. The instructions in SKILL.md and quiet-mode.md include explicit security safeguards, such as requiring user approval for destructive actions and ensuring that 'quiet' execution modes do not suppress error or block notifications. No evidence of data exfiltration, malicious execution, or harmful prompt injection was found.
Capability Assessment
Purpose & Capability
The name/description describe bounded recursive orchestration and the package contains scripts (init_run, upsert_node, mark_node, list_ready_nodes, merge_results, etc.) and schema files that directly implement that functionality. There are no unrelated env variables, binaries, or install steps requested.
Instruction Scope
SKILL.md and the scripts confine behavior to creating and manipulating a run folder (tree.json, events.jsonl, nodes/*). The guidance references using 'subagents', 'ACP', or 'direct tools' for actual leaf execution; those are runtime choices outside these scripts and could enable external actions if the agent has such tools. The scripts themselves do not perform network I/O or read arbitrary system config; they only read/write files under the chosen run directory.
Install Mechanism
No install spec is provided (instruction-only skill). Code is shipped as plain Python scripts; nothing is downloaded or extracted from untrusted URLs. Users must run the included Python scripts locally, which is low-risk if run in an appropriate directory.
Credentials
The skill requires no environment variables, no credentials, and no config paths. The scripts operate on user-supplied run directories only. There are no requests for tokens, secrets, or unrelated cloud credentials.
Persistence & Privilege
always is false and the skill does not request persistent platform privileges. It writes only into the run directory chosen by the user and does not modify other skills or system-wide agent settings.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install recursive-swarm - After installation, invoke the skill by name or use
/recursive-swarm - Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.1.1
Add quiet child execution guidance: routine child runs preserve results in session/artifacts, use ANNOUNCE_SKIP for routine announce steps, and let the parent deliver one final merged answer.
v0.1.0
Initial release: bounded recursive orchestration skill with artifact-backed task trees, events.jsonl audit trail, helper scripts, and conservative defaults.
Metadata
Frequently Asked Questions
What is Recursive Swarm?
Bounded recursive orchestration for complex tasks that are too large for one agent turn but cleanly decompose into a few independent subproblems. Use for mul... It is an AI Agent Skill for Claude Code / OpenClaw, with 277 downloads so far.
How do I install Recursive Swarm?
Run "/install recursive-swarm" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Recursive Swarm free?
Yes, Recursive Swarm is completely free (open-source). You can download, install and use it at no cost.
Which platforms does Recursive Swarm support?
Recursive Swarm is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Recursive Swarm?
It is built and maintained by Pedro Gonzalez (@plgonzalezrx8); the current version is v0.1.1.
More Skills