article-comic
Warn
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The scripts
scripts/comic.mjsandscripts/doctor.mjsuse dynamic imports (await import(...)) to load logic from a path computed at runtime. The path is derived either from the environment variableARTICLE2MEDIA_ENGINEor a relative path../../article2media/engine/scripts/. This allows the execution of code from locations that are not fixed and could be influenced by the environment. - [CREDENTIALS_UNSAFE]: The documentation in
references/mode-comic.mdindicates that the skill reads configuration from~/.config/article2media/config.jsonto manage API keys for image generation services like OpenAI and Gemini. While this is a standard location for configuration, it involves accessing a sensitive file path that likely contains secrets. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external content (articles provided as files, links, or transcripts) to generate a structured
comic.json. This structured data then drives shell-based scripts and external AI image generation services. - Ingestion points: Reads external articles from
article.mdor via upstream skills likearticle-fetch. - Boundary markers: None identified in the provided instructions for sanitizing the input article content.
- Capability inventory: The skill executes shell scripts via
Bash, writes files to the local directory (comic.json), and performs network operations (API calls to image generation providers via the shared engine). - Sanitization: No explicit sanitization or filtering of the input article content is mentioned before it is processed into comic scenes and narration.
Audit Metadata