The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/create-microsim-todo-json-files.py is vulnerable to path traversal. It extracts a sim-id from chapter documentation and uses it to construct a file path for writing without sanitizing path characters (e.g., ../), which allows for arbitrary file writes on the host system.
[DATA_EXFILTRATION]: The screenshot capture utility described in references/screen-capture.md uses Chrome with --disable-web-security and --allow-file-access-from-files flags. If an agent is instructed to capture a screenshot of a malicious MicroSim, the visualization code could read sensitive local files and exfiltrate them due to the disabled security sandbox.
[PROMPT_INJECTION]: The skill relies on parsing project documentation to drive its logic, creating a surface for indirect prompt injection (Category 8).
Ingestion points: Documentation files in docs/chapters/*/index.md are scanned for metadata and simulation specifications.
Boundary markers: None identified in the parsing scripts.
Capability inventory: File system writes (JSON output) and execution of utility scripts based on extracted content.
Sanitization: Uses json.dump for the final output format but fails to validate or sanitize the logic-driving fields like sim_id against path traversal.

microsim-utils