The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute its core functionality scripts, specifically scripts/scaffold.py and scripts/check_training_hygiene.py, as well as test scripts like scripts/scaffold_challenge/verify.sh.
[DYNAMIC_EXECUTION]: The scripts/scaffold.py script functions as a code generator. it builds a suite of Python files (model.py, dataset.py, train.py, etc.) by interpolating parameters into predefined string templates. While this is the intended purpose of the skill, it represents dynamic code generation from templates.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect injection surface because it ingests data from external CSV manifests without sanitizing headers or filenames before embedding them into the generated code.
Ingestion points: The scripts/scaffold.py script reads a user-supplied CSV file through the --manifest argument.
Boundary markers: There are no boundary markers or validation steps to ensure that the content of the CSV headers or the filename itself does not contain executable code snippets.
Capability inventory: The skill has the ability to write files to the filesystem and execute shell commands via the Bash tool.
Sanitization: The script lacks sanitization for the id_col and MANIFEST_NAME placeholders, which are replaced by the raw column name and filename respectively. A malicious user could craft a manifest with a header like patient_id\"; import os; os.system('...') # to inject arbitrary Python code into the resulting dataset.py or train.py files.

model-scaffold