The Agent Skills Directory

[SAFE] (SAFE): No malicious code or instructions detected.
The file is a markdown reference card and does not contain executable code, obfuscation, or data exfiltration patterns.
It includes helpful privacy reminders advising users to redact PII and secrets before processing images.
[COMMAND_EXECUTION] (SAFE): Documentation of a local helper script.
The file references a local script .claude/skills/vision/scripts/image_processor.py for image tasks.
The usage patterns shown (e.g., --operation resize) are standard for agent skills and do not involve shell piping or remote code execution.
[PROMPT_INJECTION] (LOW): Indirect prompt injection surface via OCR.
As a vision skill, it is designed to ingest and process text within images (OCR).
Ingestion points: Screenshots, invoices, receipts, and photos provided by users.
Boundary markers: Not explicitly defined in this reference file.
Capability inventory: The agent can extract text, analyze charts, and process documents.
Sanitization: Not mentioned; however, this is a inherent risk of the skill's primary purpose and is classified as low severity.

vision