image-cards

Warn

Audited by Gen Agent Trust Hub on Nov 12, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The scripts scripts/cards.mjs and scripts/doctor.mjs use dynamic imports to load logic from a computed path. The path is derived from the ARTICLE2MEDIA_ENGINE environment variable or a relative directory structure (../../article2media/engine/scripts/). This allows the skill to execute code from a location determined at runtime, which is a common vector for loading malicious payloads if the environment or neighboring directories are compromised.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from article.md and content.json. Instructions in SKILL.md direct the agent to take content from a generated caption.md file and "directly paste it into the response." This creates a surface where an attacker-controlled source document could embed instructions to manipulate the agent's final output to the user.
  • Ingestion points: article.md, content.json, and the generated caption.md file.
  • Boundary markers: Absent; there are no instructions to the agent to ignore or sanitize embedded commands within these files.
  • Capability inventory: The skill uses allowed-tools: Bash to run Node.js scripts that perform file system reads and writes, and requires an external browser (Chrome) for rendering.
  • Sanitization: Absent; the skill does not specify any validation or filtering for the content extracted from processed files.
  • [COMMAND_EXECUTION]: The skill relies on executing shell commands via Node.js to perform its primary functions. While scoped to the skill's directory, these scripts facilitate the dynamic loading and execution mentioned above.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 12, 2026, 04:48 PM