The Agent Skills Directory

[CREDENTIALS_UNSAFE] (HIGH): The script scripts/generate_image.py contains a check_env_file function that recursively scans the current and all parent directories for .env files to find OPENROUTER_API_KEY. This behavior can lead to the unintended exposure of secrets from unrelated projects or higher-level directories.
[DATA_EXFILTRATION] (HIGH): The script permits arbitrary local file access via the --input parameter. Any file readable by the user can be encoded and transmitted to the external OpenRouter API, posing a high risk of data exfiltration if the agent is directed to read sensitive files (e.g., SSH keys or configuration files).
[EXTERNAL_DOWNLOADS] (LOW): The skill performs network requests to openrouter.ai, which is not on the trusted external sources list. While this is the intended service, it represents a non-whitelisted external dependency.
[PROMPT_INJECTION] (LOW): SKILL.md uses 'IMPORTANT' and 'CRITICAL' directives to control agent behavior, which are classic indicators of behavioral manipulation.
[PROMPT_INJECTION] (LOW): Indirect Prompt Injection Surface. 1. Ingestion points: prompt and --input in scripts/generate_image.py. 2. Boundary markers: Absent. 3. Capability inventory: requests.post and file I/O in scripts/generate_image.py. 4. Sanitization: Absent.

generate-image