create-team
PassAudited by ClawScan on May 14, 2026.
Overview
This instruction-only skill coherently helps create runnable Claude Code agent teams, but users should review generated agents, tool permissions, hooks, and cleanup steps before using them.
Before using a generated team, review the generated subagent files, prefer project-scoped `.claude/agents/` over user-wide installation, keep tool allowlists and owned paths narrow, inspect any hooks before enabling them, and make sure you shut down teammates and delete the team when done.
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.
A generated teammate with broad tools could modify files or run commands if the user grants those tools.
Generated subagents can be given powerful tools such as Bash, Edit, and Write. This is expected for a runnable agent-team generator, but users should keep each role’s tool allowlist narrow.
`{TOOLS_ALLOWLIST}` — comma-separated tool names (Read, Grep, Glob, Bash, Edit, Write, ...)Review every generated `agents/<role>.md` file, restrict tools to the role’s real needs, and require plan approval for risky roles.
If enabled, hooks can run local commands during agent-team operation.
The templates show optional Claude Code hooks that run shell scripts on team events. This is disclosed and purpose-aligned as a quality gate, but it is still event-triggered command execution.
"TeammateIdle": [{"command": "{SKILL_ROOT}/hooks/teammate-idle.sh"}], "TaskCompleted": [{"command": "{SKILL_ROOT}/hooks/task-completed.sh"}]Inspect and customize hook scripts before wiring them into `.claude/settings.json`; avoid enabling hooks from unreviewed generated teams.
A generated team may continue coordinating and working without constant human review unless the user configures approvals and stops it cleanly.
The generated teams are intended to run with autonomous teammates. The artifacts also include teardown and plan-approval guidance, so this is disclosed rather than hidden behavior.
All members are Claude Code teammates (no human-in-the-loop required by default).
Use project-scoped installs first, set plan approval for high-impact roles, monitor spawned teammates, and follow the teardown steps when finished.