Outcome#
You can separate code collaboration from shared datasets, design group inheritance for approved data directories, and avoid destructive permission workarounds.
Concept#
A shared Git working tree mixes branches, indexes, uncommitted files, and file ownership between users. Shared datasets, checkpoints, logs, and results need a different permission and lifecycle design.
Worked Example#
Collaborators have the minimum required access, shared data has an owner, and each developer keeps a separate Git clone.
A correct example uses these decisions:
- How should several students collaborate on code? Each uses a separate clone and collaborates through branches and PRs.
- What is the safe response to a shared-folder permission problem? Inspect owner, group, ACL, and intended boundary before making a narrow change.
Common Trap#
Using one shared Git working tree or broad recursive chmod commands to solve collaboration problems.
If Blocked#
Do not guess numeric group IDs or apply broad recursive commands. Ask the storage owner to inspect the smallest affected directory. Use Euler storage for the canonical setgid/default ACL procedure when Euler is the approved system.
Useful references:
Understand Before Accepting AI Output#
An agent must not apply recursive permission changes based only on a pasted path. Verify system, owner, group, inheritance, existing contents, and rollback before any real change.