Review an AI worker’s effective file ACL before access

Select one identity and one operation before reading permissions

A trusted preprocessing worker needs to read one already admitted note F-A. A reviewer sees a named ACL entry and concludes that the worker can read it. Another reviewer cites a restrictive process umask. Neither isolated observation supplies this worksheet’s complete access decision.

Select an ordinary regular file on a supported local Linux POSIX-ACL filesystem. Independently established worker effective identity U-R differs from file owner U-O and matches the selected named-user entry. The proposed operation is reading, with no privileged override. These are explicit scope requirements, not identities derived from an agent’s job label.

Actual filesystem support, complete valid ACL, stable ownership and namespace, parent traversal, credential association, security-policy compatibility and current application rights remain unknown. They block a real positive. Owner, group-selection, other-user and privileged branches are outside this narrow calculation, not silently approved alternatives.

Keep the named entry and its mask together

The Linux ACL manual, dated 5 June 2026, checks a matching named-user entry against the process’s effective user ID. In that branch, both the entry and ACL_MASK must contain the requested permission; otherwise access is denied. When a mask exists, the group permission bits correspond to that mask. Changing corresponding mode bits also changes associated ACL entries.

This describes a discretionary permission branch, not a complete secure-file service. The anchored-open guide still owns pathname and retained-object assumptions. Its successful resolution would not replace current application authorization, and this permission worksheet would not replace its object checks.

Keep a protected review record identifying the exact file revision, admitted worker context, complete policy revision and intended operation. A remembered display or a copied entry can become stale. The actual coordination that keeps those associations stable through an operation is unimplemented here; an earlier review alone cannot close that timing gap.

Use a small effective-operation sheet

The following permission words are invented labels, not an encoded ACL, valid configuration or host inspection. The table deliberately omits the rest of the ACL, which still requires independent complete review. It demonstrates one selected calculation without supplying a grant command.

The useful conditional positive uses the first row: the complete actual policy and every independent prerequisite permit U-R’s intended read of F-A. It admits that bounded plan for separately authorized implementation review. It reports no opened file, captured bytes or successful model task.

The last two rows explain why reviewing a currently restricted effective result is not the same as approving every later policy change. Our proposed governance rule records broader dormant entries as review concerns and requires new review when the effective mask changes. It does not prescribe a particular ACL layout or claim the documentation forbids broader entries.

Review creation inheritance separately from current access

The umask manual, Linux man-pages 6.19 dated 8 February 2026, distinguishes the ordinary creation-mask path from creation under a directory’s default ACL. In the latter case, umask is ignored: the default ACL is inherited, and permissions absent from the creation mode are removed. A process-mask setting by itself therefore cannot establish the new file’s resulting access policy.

This is a second review question, not a file-creation recipe. Before accepting a claim such as “our outputs are private because the service sets umask,” require supported evidence about the actual parent context and requested creation mode. Do not assume the same setting describes existing files, another directory or a different filesystem.

A creation record and a current-access record serve different purposes. The former explains how an initial policy should arise; the latter must establish the policy and identity applicable now. Later changes, copied artifacts and uncontrolled lifecycle paths remain outside the worksheet’s evidence. No inheritance setting was applied or inspected for this article.

Walk eight inert access-review cards

Separate a permission review from a deployed boundary

A mode-looking number does not identify the file, authenticate the worker or prove its bytes stayed unchanged. Likewise, removing unnecessary environment credentials does not settle every filesystem grant. Keep namespace, descriptor ownership, worker behavior, current data rights and output handling in their existing independent reviews.

No candidate ACL, file, directory, identity, permission, capability, descriptor or process configuration was created, inspected or changed. No access check, file creation, permission mutation, host probe, fixture or model runtime was executed. Ordinary editorial file handling is separate. The worksheet implements no credential-drop, ACL-management or revocation service.

These two actual manual bodies establish the documented scope only, not an installed kernel, library or filesystem. Boundary remains prelaunch with planned request-path checks. Its synthetic policy evidence does not validate this file-access profile. Permission eligibility, useful processing and authorized disclosure remain separate claims.