The Agent Skills Directory

[SAFE]: The skill implements a robust versioning system for all file edits using the edit.sh and restore.sh scripts, ensuring no work is lost and providing a clear audit trail.
[SAFE]: All shell scripts use security best practices, including parameter expansion for usage checks, strict error handling (set -euo pipefail), and properly quoted variables to prevent command injection.
[SAFE]: Data manipulation of metadata files is handled through jq, which provides a secure way to update JSON structures without risk of injection.
[SAFE]: The skill does not perform any external network requests or download third-party code, keeping all operations local to the user's machine.
[SAFE]: Although the skill processes untrusted user data in writing pieces, the supporting scripts are designed to handle filenames and content safely.
- Ingestion points: Data enters the system via pieces/{id}/content.md and pieces/{id}/brief.md.
- Boundary markers: Standard Markdown structure is used, though no specialized delimiters are present to isolate user content from the agent's internal logic.
- Capability inventory: The skill is limited to local file system operations (create, write, delete versions) via the provided shell scripts.
- Sanitization: Variables are quoted in scripts, and metadata is managed via the jq utility to ensure data integrity.

Write