baoyu-compress-image
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [Indirect Prompt Injection] (HIGH): The skill is designed to ingest and process untrusted external data (user-provided images). Maliciously crafted images or metadata could exploit vulnerabilities in the underlying processing tools (e.g., ImageMagick's history of CVEs like ImageTragick).
- Ingestion points: The
<input>argument inscripts/main.ts. - Boundary markers: None identified in the instruction file to isolate untrusted data.
- Capability inventory: Shell command execution (sips, cwebp, ImageMagick) and file system write access.
- Sanitization: No sanitization logic is described in the prompt instructions to prevent exploitation of the image processing tools.
- [Command Execution] (MEDIUM): The skill invokes system-level binaries via a CLI wrapper. If the
main.tsscript does not properly sanitize filenames or options before passing them to the shell, it could be vulnerable to command injection. - [External Downloads] (LOW): The execution command
npx -y bunallows for the automated download of the bun runtime from the npm registry at runtime. While bun is a standard tool, the lack of version pinning or integrity checks throughnpx -yis a minor security risk.
Recommendations
- AI detected serious security threats
Audit Metadata