The Agent Skills Directory

[DYNAMIC_EXECUTION]: The archive_generator function in scripts/office_repair.py modifies existing Python scripts on disk by prepending a code guard. This guard utilizes atexit and filesystem operations to ensure file integrity during execution.
[DYNAMIC_EXECUTION]: The _bootstrap_ane_package function dynamically modifies sys.path at runtime to include paths derived from environment variables and the user's home directory (OneDrive), which is used to load the ane_package library.
[COMMAND_EXECUTION]: scripts/test_office_repair.py executes shell commands and Python scripts using subprocess.run to verify the behavior of the driver and its code-guarding mechanism.
[INDIRECT_PROMPT_INJECTION]: The skill processes Office documents (Excel, Word, PowerPoint) which serve as untrusted input. Malicious content within these files could potentially influence the agent's behavior or interact with the skill's capabilities.
Ingestion points: Office files are read using openpyxl, python-docx, python-pptx, and Excel COM interop in scripts/office_repair.py.
Boundary markers: The skill follows an "edit-preservation-protocol" and requires explicit user agreement on fix lists before applying changes.
Capability inventory: Includes file-write operations, Office COM interop (via PowerShell), and subprocess execution (in tests).
Sanitization: The diff logic identifies common Office autocorrect artefacts but does not include sanitization for embedded instructions.

office-repair