The Agent Skills Directory

[SAFE]: The skill implements significant defensive coding practices to manage filesystem access. In scripts/adr_scribe/paths.py, all path resolutions use os.O_NOFOLLOW and directory file descriptors (dir_fd) to ensure that operations do not follow symlinks, effectively preventing symlink-based path traversal attacks.- [SAFE]: Git operations are heavily hardened in scripts/adr_scribe/gitcmd.py. The skill forces a whitelist of read-only subcommands and explicitly disables dangerous features such as git hooks (core.hooksPath=/dev/null), external diff tools, and system-level configuration files to mitigate risks from malicious repository settings.- [SAFE]: A secure transaction model is used for writing records. The prepare-record script stages changes outside the repository, and the apply-record script requires an explicit user approval of the patch digest before any files are written. This ensures human-in-the-loop verification for all repository mutations.- [SAFE]: The skill includes a dedicated validation engine in scripts/adr_scribe/validate.py that automatically scans proposed content for sensitive data like credentials and tokens. It also verifies that any suggested shell commands in the documentation are read-only and conform to a safe whitelist.- [SAFE]: Persistence and locking mechanisms in scripts/adr_scribe/journal.py are implemented using atomic directory creation and a write-ahead journal. This prevents race conditions and ensures the repository remains in a consistent state even if the process is interrupted.

adr