The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The main script scripts/omni_image.py contains a hardcoded Google API key under the Google provider configuration: hardcoded_key: "AIzaSyCZqClxP5DXTSZEoE22KcbHJBKI1fPPj9o". Hardcoding secrets in source code is a major security risk as it exposes the developer's quota and access to the service.
[COMMAND_EXECUTION]: The wrapper script examples/python_wrapper_example.py uses the subprocess.run() function to execute a system command omni-image. While the provided example uses static prompts, this pattern is dangerous if user-supplied input is incorporated into the prompts without strict sanitization, potentially leading to shell command injection.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data (images and JSON manifests) and passes that data to powerful AI models.
Ingestion points: The describe command in scripts/omni_image.py analyzes arbitrary image files using vision models, and the batch command processes external JSON manifests.
Boundary markers: There are no explicit delimiters or instructions to the model to ignore potential malicious content embedded within images (e.g., text-in-image attacks).
Capability inventory: The skill possesses the ability to make network requests to multiple AI providers and write files to the local disk.
Sanitization: The script concatenates prompt segments (subject, style, etc.) and style prefixes without any sanitization or validation to filter out adversarial instructions.
[EXTERNAL_DOWNLOADS]: The unified provider in scripts/omni_image.py points to https://api.302.ai/v1. This is a third-party API aggregator/proxy service. Sending API requests and keys through a third-party intermediary introduces a risk of data interception or service manipulation by the proxy provider.

omni-image