The Agent Skills Directory

[COMMAND_EXECUTION]: The get_article_summary function in scripts/consult.sh is vulnerable to Python code injection. It uses shell variable interpolation to place the $slug argument into a python3 -c command string. This allows for arbitrary Python code execution if the argument is maliciously crafted by providing a string that closes the Python function and executes system commands.\n- [COMMAND_EXECUTION]: The search_related and search_and_summarize functions in scripts/consult.sh are vulnerable to argument injection. The $query variable is passed directly to the grep command without sanitization or the use of -- to terminate options, which can be used to inject malicious flags to read unintended files or modify command behavior.\n- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection.\n
Ingestion points: Article data is ingested from JSON files located in ~/drli_articles/ via the consult.sh script.\n
Boundary markers: There are no clear delimiters or instructions to ignore embedded commands when the article content is printed for the agent.\n
Capability inventory: The skill facilitates file reading and Python execution via its consulting script.\n
Sanitization: The script does not perform any sanitization or escaping of the content retrieved from the article data files before displaying it to the agent context.

binlee-advisor