The Agent Skills Directory

Indirect Prompt Injection Surface: The utility scripts/synthetic_data_operation.py processes external JSON records to authorize and execute file operations, which represents a potential surface for indirect prompt injection.
Ingestion points: The script reads JSON preflight and result records from user-specified file paths.
Boundary markers: The implementation processes structured JSON data but does not employ specific boundary markers to isolate input data from potential instructions within the agent's context.
Capability inventory: The utility has the capability to read, copy, and replace local files, though these actions are restricted by internal logic.
Sanitization: The script performs structural validation using jsonschema and verifies file integrity via SHA-256 digests before proceeding with file modifications.
Controlled File System Interaction: The commit_local function in scripts/synthetic_data_operation.py performs file replacements and creates backups.
Containment Measures: The script implements path resolution and containment checks to ensure that all file operations are restricted to an approved root directory, mitigating path traversal considerations.
Safety Mechanisms: It uses atomic file replacement via os.replace and performs pre-commit digest re-validation to ensure the target file has not changed unexpectedly.
External Dependency Management: The skill defines several third-party Python dependencies in its configuration files.
Package References: Dependencies include jsonschema for data validation and development tools such as pytest, ruff, and the atheris fuzzing library.
Source Verification: All identified packages and their specific versions in uv.lock originate from official package registries and are consistent with the skill's stated purpose for testing and validation.

dataops