The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow script (run_workflow.py) orchestrates multiple tasks by calling local Python scripts using the subprocess module. These calls pass arguments as lists, avoiding shell injection risks. The scripts being called are internal to the skill or linked components.
[DATA_EXFILTRATION]: The skill reads preferences from a specific configuration path in the user's home directory (~/.alicloud-skills/). This is a legitimate and standard practice for persistent configuration storage, used here to preserve user settings for the illustration workflow.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied Markdown articles to determine suitable illustration topics.
Ingestion points: The source.md file provided via the --source argument.
Boundary markers: The workflow requires an outline.md confirmation step and batch confirmation of settings before generation.
Capability inventory: File system writes (artifacts), subprocess execution (internal helpers and image generators).
Sanitization: The skill extracts textual information into structured prompt templates but does not perform explicit sanitization of Markdown control characters or prompt instructions within the article content, relying on the agent's summary capabilities.

aliyun-solution-article-illustrator