Is shadcn-ui/ui safe?
Yes. shadcn-ui/ui is safe to install. Oathe's behavioral security audit gave the ui skill by shadcn-ui a trust score of 90/100 with 5 findings, none critical or high.
https://github.com/shadcn-ui/ui
Is shadcn-ui/ui safe to install?
This is the official shadcn/ui repository from the shadcn GitHub organization — a widely-used, legitimate React UI component library with two agent skill definitions (shadcn and migrate-radix-to-base). Both SKILL.md files contain professional development tooling with no adversarial prompt injection patterns, hidden instructions, or data exfiltration attempts. Credential file accesses observed during monitoring are consistent with the monitoring infrastructure's honeypot baseline verification rather than repository code, as the git clone failed before any install scripts could execute and all canary files remain intact. The primary residual risks are standard supply chain exposure from npm package execution and the broad but documented agent capabilities the skill grants.
What security issues were found in shadcn-ui/ui?
Category Scores
Findings (5)
LOW Live command execution on skill load -10 ▶
The shadcn SKILL.md contains !\npx shadcn@latest info --json`` which executes a remote npm package at skill-load time to inject project context. This is documented intentional behavior in the shadcn skill system but means loading the skill downloads and runs code from the npm registry (the shadcn package), creating a supply chain trust dependency.
Evidence: apps/v4/skills/shadcn/SKILL.md line: `!\`npx shadcn@latest info --json\``
LOW Skill grants significant filesystem modification capability -5 ▶
The shadcn SKILL.md defines allowed-tools: Bash(npx shadcn@latest *) which permits the agent to run any shadcn CLI subcommand. This includes commands that install components, modify globals.css, tsconfig.json, and package.json. This is intentional and documented but represents meaningful write access to the user's project files.
Evidence: SKILL.md frontmatter: `allowed-tools: Bash(npx shadcn@latest *), Bash(pnpm dlx shadcn@latest *), Bash(bunx --bun shadcn@latest *)`
INFO Credential files accessed during monitoring window -9 ▶
inotify and auditd recorded OPEN/ACCESS/CLOSE_NOWRITE operations on .env, .ssh/id_rsa, .aws/credentials, .npmrc, .docker/config.json, and .gcloud credentials at timestamp 1794816444.842. All six accesses occurred within <1ms, all were read-only, canary file integrity is confirmed intact, and the timing aligns with SSH session establishment — consistent with the monitoring infrastructure performing its honeypot baseline verification rather than repository-originated activity. No repository code or install scripts executed (clone failed).
Evidence: auditd PATH records 26585-26590 at timestamp 1794816444.842; inotify events at 08:06:58; install output: 'fatal: destination path already exists'
INFO Git clone did not execute — directory pre-existed -10 ▶
The git clone command failed because the skill directory was already present in the sandbox. This means no install scripts, postinstall hooks, or other repository-triggered code ran during the audit. The monitoring captured file accesses from the pre-staged repository and SSH infrastructure only.
Evidence: Install output: `fatal: destination path '/home/oc-exec/skill-under-test/shadcn-ui/ui' already exists and is not an empty directory.`
INFO Skill grants broad agent context-awareness via live project introspection -10 ▶
The shadcn skill injects live project state into the agent context via CLI execution on load, then provides detailed instructions guiding agent behavior across component installation, preset management, and design system creation. The instructions are legitimate and transparent but are extensive — this is a high-capability skill that significantly shapes agent behavior during shadcn-related tasks.
Evidence: SKILL.md sections: 'Current Project Context', 'Critical Rules', 'Workflow', spanning ~300 lines of behavioral guidance
Should I install shadcn-ui/ui?
Oathe's verdict for shadcn-ui/ui is SAFE with a trust score of 90/100. Recommendation: Install.