Skip to content

gate-57's foundation guard keys on the checkout directory name, so it never engages in CI (and gate-7's $this-> sanitiser false positives) #398

Closed

Description

@rubenvdlinde

gate-57's foundation guard is keyed on the checkout's directory name — so it has never engaged in CI, for any repo

quality.yml checks out with path: app. gate-57's hydra#106 fail-safe identifies the foundation repo by directory name, so under CI the directory is always app and the guard never matches.

Four-arm control, same openregister tree, only the directory name varying:

directory gate-57 findings
openregister 0
renamed 8

So every gate-57 figure ever taken from a CI run is a measurement of the guard not engaging, and every figure taken from a local clone named after its repo is a measurement of it engaging. The two are not comparable, and neither is labelled.

Corrected figures for the current fleet-impact table: openregister 8, openconnector 3.

🔑 A guard keyed on a path component is keyed on something the caller controls and the checker cannot see. The identity should come from something intrinsic — appinfo/info.xml's app id, or the repository name from the environment — not from where the tree happens to sit.

Related, and more urgent for anyone about to act on gate-7 findings

gate-7 fires on any correctly-scoped method that sanitises a parameter through a $this-> helper. Isolated in both directions:

  • adding $safeKey = $this->sanitizeKey($key) flips a clean method to FINDING
  • adding a session identity to that helper call clears it

This is a systematic false-positive class, not a per-app accident. Adding an authorisation guard would be the wrong repair — the method is already correctly scoped, and a guard would be dead code written to satisfy a misreading.

⚠️ Anyone triaging the +411 fleet-wide gate findings from the #373/#376/#377/#378/#379/#380 wave must be briefed on this before touching gate-7. The size of the class is currently unquantified — a first screening attempt was invalid (it matched the 401 preamble itself) and was discarded rather than reported.

Context: what the wave actually did to the fleet

Measured OLD 487e8d23 vs NEW 97165ed4, both --full, no delta base, each app's current development:

6205 → 6616 findings, +411, with zero application-code change.

Every app sees this on its next run — all 18 callers use quality.yml@main with enable-hydra-gates: true and no hydra-gates-ref pin. There is no release to wait for.

Nearly all of the +411 is pre-existing debt now visible rather than anything new. Verified on a real case: in docudesk's same file, batchUpload() (passes a session userId) is correctly suppressed while folderBatch() (caller-supplied folderId, only a 401 preamble) is reported. That is the gate working.

The two exceptions are the two above.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions