The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/render.py to execute Chromium for rendering HTML character sheets. The command is invoked with the --no-sandbox flag, which reduces browser security isolation. Furthermore, the command arguments include file paths and file:// URLs constructed from character IDs (ficha["id"]); if these IDs are not properly sanitized by the agent, they could lead to path traversal or writing artifacts to unintended directories.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to a local image generation server at http://localhost:8000 (scripts/imgclient.py). It also contains an optional quality control feature in scripts/qc_imagem.py that sends generated images and prompt data to the Anthropic API (https://api.anthropic.com) for vision-based evaluation. Both are functional requirements and target local or trusted endpoints.
[INDIRECT_PROMPT_INJECTION]: The skill accepts user-provided character descriptions and histories which are interpolated into prompts for image generation models and HTML templates. While HTML output is escaped, the image generation prompts are constructed by simple string concatenation.
Ingestion points: The character data (ficha.json) derived from user input or history.
Boundary markers: Absent; user-provided descriptions are integrated directly into generation prompts in scripts/artes.py.
Capability inventory: System command execution (Chromium), network communication, and local file system writes.
Sanitization: Uses html.escape for character sheet rendering, but lacks robust sanitization for the instructions sent to the image generator.

inemaref-folder