The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project content, creating a potential attack surface. 1. Ingestion points: The scripts/find_duplicate_ui.py script reads .vue files using rglob. 2. Boundary markers: No explicit boundary markers or 'ignore' instructions are used during file reading. 3. Capability inventory: The main script performs only file reads and JSON output; the shared _skill_cli.py contains subprocess.run capabilities but these are not utilized by the duplicate finder tool. 4. Sanitization: The script tokenizes template content to calculate Jaccard similarity and only outputs filenames and numerical scores. Since raw content is not included in the tool response, the injection risk is neutralized.
[COMMAND_EXECUTION]: The scripts/_skill_cli.py library includes a run_command helper using subprocess.run. It follows security best practices by setting shell=False and using shutil.which to resolve absolute paths. This utility is not called by the active tool script.
[SAFE]: The skill operates entirely on local UI files to provide architectural feedback and exhibits no malicious behaviors such as data exfiltration, obfuscation, or persistence.

no-duplicate-ui