The Agent Skills Directory

[SAFE]: The skill implements a robust reset mechanism that prioritizes data integrity. The reset.py script uses a SHA-256 fingerprinting technique to verify that the project state has not changed between the user's preview and the actual reset execution. It also performs atomic file replacements using temporary directories and os.replace, ensuring that the system is not left in a corrupted state if the process is interrupted.
[COMMAND_EXECUTION]: The skill executes a local Python script (reset.py) to manage file operations. The script is restricted to reading and writing within a designated state_directory and only modifies a specific set of markdown notes (profile.md, progress.md, project-map.md). It includes validation to ensure the provided project path is an absolute directory and exists on the filesystem.
[INDIRECT_PROMPT_INJECTION]: The skill accepts a project directory path as input, which represents a potential injection surface. However, this risk is effectively mitigated by explicit instructions in SKILL.md directing the agent to safely quote the path when calling the shell. Furthermore, the reset.py script uses the argparse library and pathlib validation, which provides defense-in-depth against malicious path strings at the execution level.

reset