The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file SKILL.md contains hardcoded plaintext credentials for an external media server in Phase 6. The username delorenj and password Ittr5eesol are explicitly defined within a Python code snippet intended for asset uploads to https://media.delo.sh.
[DATA_EXFILTRATION]: The skill instructs the agent to source environment variables from a non-standard local path: ~/.config/zshyzsh/secrets.zsh. Accessing and executing scripts that load local secret files increases the risk of credential exposure or unintended access to sensitive information.
[COMMAND_EXECUTION]: The workflow relies on several shell and Python commands to process data. This includes a custom Python script for image cropping in Phase 2, a standalone conversion script scripts/coloring-convert in Phase 3/4, and ImageMagick (magick) for file format conversion in Phase 6.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted visual data from Google Images in Phase 1 which is then processed by a sub-agent for quality assurance in Phase 5.
Ingestion points: Browser screenshots of external search results (Phase 1/2).
Boundary markers: No delimiters or safety instructions are provided to the sub-agent to ignore embedded instructions in the images.
Capability inventory: The skill possesses Python execution capabilities (Phase 2, 7), network upload access (Phase 6), and browser-based publishing tools (Phase 9).
Sanitization: No sanitization or validation of the external image content is performed before processing.
[EXTERNAL_DOWNLOADS]: The skill downloads generated assets from fal.ai storage after processing. These interactions with well-known technology services are documented as part of the intended pipeline functionality.

creating-coloring-books