Static inspection shows high-risk AI-agent configuration and installer commands, but no unconsented lifecycle execution, credential harvesting, covert exfiltration, persistence outside the project config, or destructive behavior. Because the dangerous primitives are interactive and package-aligned, this should not be publish-blocked.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
User runs `start-vibing-stacks`/`svs` and confirms setup, optionally selecting MCPs and allowing installs.
Impact
A configured project grants Claude Code broad local tool permissions and hook execution; missing dependencies may be installed from public installer URLs.
Mechanism
interactive CLI writes Claude Code settings/hooks and launches Claude with skipped permissions
Policy narrative
No hidden install-time attack was found. The CLI deliberately bootstraps Claude Code workflows, writes project `.claude` hooks/settings, may install dependencies, and then launches Claude with `--dangerously-skip-permissions`; that is risky but visible, user-invoked, and consistent with the package description.
Evidence
package.jsondist/index.jsdist/setup.jsdist/installer.jsdist/mcp.jsstacks/_shared/config/mcp-servers.json