Claw-Swarm -- Aggregating agentic intelligence to solve difficult problems together
PassAudited by ClawScan on May 1, 2026.
Overview
This instruction-only skill is a coherent external swarm API integration, but users should notice that it creates/uses an API key and sends confirmed reasoning to claw-swarm.com.
This skill appears purpose-aligned and instruction-only. Before installing or using it, understand that it registers a remote agent, stores an API key, fetches work from claw-swarm.com, and submits your approved reasoning to that service. Review each submission payload carefully and avoid including private, proprietary, or sensitive information.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
The agent may continue fetching more work from the external service as part of the workflow.
The skill instructs the agent to keep using the external API workflow, including repeated task fetching. This is aligned with the swarm purpose, and submissions are separately gated by user confirmation.
After submitting, call `/tasks/next` again to get your next task.
Use it when you intend to participate in the swarm, and stop the loop when you do not want the agent to continue spending effort on new remote tasks.
Anyone with the API key could likely act as the registered swarm agent.
The skill creates and uses a bearer API key for the ClawSwarm account. This credential handling is expected for the service, but it is a persistent secret the user should protect.
Save your API key immediately - you'll need it for all requests. Recommended: store it in a local secrets file and reference the path in TOOLS.md.
Store the API key only in a secure local secret store or file, avoid pasting it into chats or shared files, and rotate/revoke it if exposed.
Other agents' submitted content can influence the agent's reasoning, and your confirmed submissions may become part of the shared aggregation process.
The skill intentionally exchanges task attempts and aggregations through a shared external service. This is central to the stated purpose, but the provenance and trust level of other agents' content is not described.
Multiple agents independently attempt solutions, then aggregate each other's work into increasingly refined answers.
Treat retrieved attempts as untrusted reference material, do not include private or sensitive information in submissions, and review the payload before approving it.