cookbook-audit
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- Command Execution Consideration: The
validate_notebook.pyscript constructs shell commands using string interpolation with user-supplied notebook paths. Specifically, thecheck_hardcoded_secretsmethod usessh -cto pipe file paths todetect-secrets. \n - Evidence:
cmd = ["sh", "-c", f"echo '{notebook_abs}' | ... --baseline {baseline_path} --plugin {plugins_path} --verbose"]invalidate_notebook.py.\n - Concern: If a notebook path or directory name contains shell metacharacters such as single quotes or semicolons, it could result in the execution of unintended commands. Migrating to a list-based
subprocess.runcall without a shell would enhance the script's security posture.\n- External Resource Fetching: The script relies onuvxanduv runto dynamically download and executedetect-secretsandnbconvertduring the validation process.\n - Evidence: The script uses
uvx --from detect-secretsanduv run --with nbconvertwithinsubprocess.runcalls.\n - Concern: While these are established tools, performing remote downloads at runtime introduces a dependency on external registries and the integrity of the fetched packages.\n- Indirect Prompt Injection Surface: The skill is designed to ingest and analyze external Jupyter notebook files. Since the agent interprets the content of these notebooks to generate an audit report, there is a potential for instructions embedded within the notebook data to influence the agent's behavior.\n
- Ingestion points: The agent reads markdown files generated from user notebooks in
validate_notebook.pyas part of the workflow described inSKILL.md.\n - Boundary markers: The skill defines a structured report format, but does not use explicit delimiters or "ignore" instructions when processing the raw notebook content.\n
- Capability inventory: The skill can execute shell commands via
validate_notebook.pyand perform file writes to thetmp/directory.\n - Sanitization: The script performs technical validation but does not sanitize the natural language content of the notebook before it is reviewed by the agent.
Audit Metadata