The Agent Skills Directory

[PROMPT_INJECTION]: The scripts/gen.py script is vulnerable to indirect prompt injection via a Cross-Site Scripting (XSS) vulnerability. User-controlled prompts are interpolated directly into the <figcaption> tags of the generated index.html gallery without any sanitization.
- Ingestion points: args.prompt argument in scripts/gen.py
- Boundary markers: None present in the write_gallery HTML template
- Capability inventory: File-write (index.html) and network-read (urllib.request)
- Sanitization: No HTML escaping or sanitization is applied to the prompt string before output.
[PROMPT_INJECTION]: The skill uses SKILL.md metadata and a separate SAFETY.md file to make self-referential claims about its own safety ('Armored by Faberlens', 'Verdict: PASS'). This is a deceptive pattern intended to influence the safety evaluation of the agent or analyzer (Category 8e).
[COMMAND_EXECUTION]: Executes the local Python script scripts/gen.py to orchestrate the image generation and gallery creation workflow.
[EXTERNAL_DOWNLOADS]: Fetches generated image content from OpenAI's official API endpoints using the urllib.request.urlretrieve method to save them to the local filesystem.

openai-image-gen-hardened