The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/csv-library.py uses importlib.util to dynamically load a Python module from a path computed relative to its own location (../../../../tools/workspace/product_library.py). This allows the skill to execute code that is not contained within the skill's package, which represents a significant risk if the workspace directory structure can be influenced by an attacker.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, which could be used to deliver malicious instructions to the agent.
Ingestion points: Untrusted data is read by scripts/csv-library.py from files provided via the --source (CSV) and --row-json (JSON) arguments.
Boundary markers: The instructions in SKILL.md define a manual 'native mutation sequence' for the agent to follow (Inspect, Prepare, Verify, Apply), but there are no automated boundary markers or specific instructions to ignore embedded commands within the processed data.
Capability inventory: The skill is granted access to file system operations and shell command execution (Bash) through the host's platform configuration.
Sanitization: The script relies on standard csv and json libraries for parsing but lacks content-level sanitization or escaping to prevent the interpolation of malicious data-embedded instructions into the agent's reasoning process.

product-library