Swarm
ReviewAudited by ClawScan on May 10, 2026.
Overview
The skill is mostly aligned with parallel research, but it grants broad shell access and can search local knowledge, past sessions, and project code without clear limits.
Use this skill only if you are comfortable with a multi-agent research workflow. Before running it, ask the agent not to use Bash unless explicitly approved, and specify whether it may search local knowledge bases, past sessions, or project code.
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.
During a research task, the agent could run local shell commands beyond what is needed to gather and write research findings.
The skill grants Bash shell access even though the documented workflow is web/local research and writing a report; no scoped Bash commands or limits are provided.
allowed-tools: Read, Grep, Glob, Bash, WebSearch, WebFetch, Write, ...
Remove Bash from allowed tools, or restrict it to specific non-destructive setup actions such as creating the docs directory and require explicit user approval.
Private prior conversations, notes, or codebase details could be incorporated into the generated research report or influence the recommendation.
The skill can pull from past sessions, a local knowledge base, and project/code indexes, but does not bound which sessions, projects, paths, or sensitive data should be excluded.
If MCP `session_search` available: check if this idea was researched before in past sessions ... MCP `kb_search` ... search local knowledge base ... `project_code_search` ... search for reusable patterns, services, infrastructure across existing projects
Ask the user before using local knowledge, past-session, or project-code searches; limit searches to named projects or paths; and avoid copying sensitive local content into research.md.
Information found by one teammate, including local or sensitive context if enabled, may be shared with the other teammates and included in the final synthesis.
The skill intentionally uses multiple agents sharing findings. This is central to the purpose, but the artifacts do not define separate data boundaries for each teammate.
Spawn 3 teammates ... Each teammate writes findings to a shared task list
Keep the plan-approval step and make the plan state which data sources each teammate may use, especially for local or private project information.