The Agent Skills Directory

[COMMAND_EXECUTION]: The validate_notebook.py script contains a shell in jection vulnerability in the check_hardcoded_secrets method. It interpolates the user-provided notebook path into an f-string used as a shell com m an d: f"echo '{notebook_abs}' | ...". An attacker can provide a path with a single quote to escape the string an d execute arbitrary com m an ds via the sh -c execution context.
[REMOTE_CODE_EXECUTION]: The skill's validation script dynam ically downloads an d executes the detect-secrets an d nbconvert packages using uvx an d uv run. This remote execution pattern is highly risky when paired with the com m an d in jection vulnerability, as it provides a pathway for execution of arbitrary logic from external sources.
[COMMAND_EXECUTION]: The convert_to_markdown function invokes jupyter nbconvert using subprocess.run. This exposes the agent to the potential vulnerabilities of a com plex docum ent conversion tool when processing untrusted user-provided notebook files.

cookbook-audit