Is ui-audit safe?
https://clawhub.ai/tommygeoco/ui-audit
The ui-audit skill is a well-structured, content-focused UI/UX auditing framework based on the published book 'Making UX Decisions' by Tommy Geoco. It contains no malicious code, no data exfiltration mechanisms, no prompt injection attempts, and no suspicious runtime behavior. The only executable component is a benign postinstall echo statement. All 20+ reference files contain purely educational UX pattern documentation.
Category Scores
Findings (5)
LOW Postinstall script in package.json -8 ▶
The package.json includes a postinstall script that executes an echo command. While this specific instance is benign (static informational text only), postinstall scripts are a common vector for supply-chain attacks. The script contains no variable interpolation, command substitution, or network calls.
LOW Structured output format directive -5 ▶
The skill instructs the agent to generate audit reports in a specific JSON format with defined fields. This is a legitimate templating pattern for a UI audit skill, but it does shape agent output behavior. The format does not include any fields that would capture sensitive system information.
INFO External URL references in metadata -2 ▶
The skill references external URLs (https://audit.uxtools.co, https://uxdecisions.com) as attribution and homepage links. These are not fetch directives — they appear in YAML frontmatter and markdown links for informational purposes only. The agent is not instructed to visit or fetch content from these URLs.
INFO Includes a CLAUDE.md file -5 ▶
The skill ships a CLAUDE.md file which, if copied to a project root as suggested by the postinstall message, would inject instructions into Claude Code's system context. This is a documented and expected pattern for Claude Code customization, but users should review the contents before copying.
INFO Installer runtime reads environment files -8 ▶
The filesystem monitor recorded OPEN/ACCESS events on /home/oc-exec/.env, .aws/credentials, .openclaw/openclaw.json, .profile, and .bashrc during installation. These reads are attributable to the openclaw gateway/installer process initialization (consistent timestamps, standard shell init sequence), not to the skill code itself. No writes were detected on any of these files.