The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local JSON manifests and CSV transfer tables via scripts like validate_manifest.py and plan_transfers.py. It effectively mitigates potential injection attacks through a comprehensive validation framework in _common.py that includes:
Ingestion points: Files are read using load_json and load_csv from paths restricted to the current working directory via safe_input_path.
Boundary markers: The skill enforces a strict JSON schema (protocol-manifest.schema.json) and fixed CSV header formats.
Capability inventory: The processing scripts are restricted to data transformation and stdout reporting; they lack capabilities for network operations, file system modifications, or subprocess execution.
Sanitization: Inputs undergo rigorous type checking, numeric range validation (e.g., MAX_VOLUME_UL), and regular expression matching (e.g., WELL_RE, NAME_RE).
[DYNAMIC_EXECUTION]: The script inspect_backends.py performs lazy loading of PyLabRobot modules to verify API compatibility. This dynamic loading is restricted to a fixed list of known symbols within the specific version of the library and does not involve untrusted computed paths or user-supplied execution strings.
[EXTERNAL_DOWNLOADS]: The skill documentation provides instructions for installing the PyLabRobot package from PyPI. This is a legitimate dependency for the skill's functionality, targeting a recognized open-source library.
[DATA_EXPOSURE]: The skill implements strong safeguards against path traversal in its safe_input_path helper, which explicitly resolves paths to ensure they remain within the current working directory and forbids the use of symlinks to access files outside the project scope.

pylabrobot