creating-coloring-books
Fail
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill contains a plain-text username and password for a media gallery service used during the asset upload phase.
- File: SKILL.md
- Evidence:
username=delorenj, password=Ittr5eesol - [DATA_EXFILTRATION]: The skill instructions and scripts access a local ZSH configuration file to load environment secrets and API keys.
- File: SKILL.md, scripts/coloring-convert
- Evidence:
source ~/.config/zshyzsh/secrets.zshand reading of theFAL_KEYenvironment variable. - [COMMAND_EXECUTION]: The skill executes local Python scripts to perform image conversion and uses sub-agent sessions for quality assurance tasks.
- File: SKILL.md
- Evidence: Execution of the
coloring-convertscript and use of thesessions_spawntool. - [INDIRECT_PROMPT_INJECTION]: The skill captures screenshots from external search engine results and processes them through AI models without sanitization, creating an attack surface where adversarial image content could influence the agent's behavior or the published output.
- Ingestion points: Web screenshots of Google Image search results (Phase 1 & 2).
- Boundary markers: None present.
- Capability inventory: Image processing via external fal.ai APIs, file uploads to media.delo.sh, and automated browser-based listing creation on storefronts like Etsy, Creative Market, Gumroad, and Amazon KDP.
- Sanitization: The skill implements image cropping for extraction purposes but lacks content filtering or safety validation.
Recommendations
- AI detected serious security threats
Audit Metadata