The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill is designed to ingest and process untrusted external content from MicroSim directories.
Ingestion points: Reads user-supplied main.html, index.md, and *.js files within the docs/sims/ directory structure for validation and indexing.
Boundary markers: Absent; the skill lacks delimiters or instructions to treat MicroSim content strictly as data, increasing the risk of the agent obeying instructions embedded in the analyzed code.
Capability inventory: Possesses significant side-effect capabilities, including writing PNG files, generating icons, and modifying the project's mkdocs.yml navigation structure.
Sanitization: Absent; no sanitization or validation of the external content is described before it is processed by the agent or automation tools.
[Command Execution] (MEDIUM): The skill documentation explicitly mentions using 'Chrome headless mode' for 'screen-capture.md'. Execution of browser automation against untrusted local HTML/JS is a known vector for local file access or further exploitation if the browser environment is not strictly sandboxed.
[Unverifiable Logic] (MEDIUM): The core operational logic for each utility is stored in external files (e.g., references/standardization.md, references/screen-capture.md). The main SKILL.md acts only as a router, meaning the exact commands and safety boundaries implemented in those reference files are unverified in this analysis.

microsim-utils