← Pilot offer

Independent MCPB regression proof

Internal sample executed on Linux, October 10, 2026. No customer order or payment.

Recorded result

Both upstream file walkers were exercised. Controls covered intact sibling contents, valid file symlink contents and target permissions, valid directory aliases and explicit ignore rules. Broken-link ignore count was checked where exposed by the API.

Attribution

The fix is the existing upstream author's work, not ours. See issue #300 and PR #303. The recorded run used actual pinned upstream source, not a reimplementation.

Limits

Function-level regression evidence on Linux only. Not a full MCPB CLI build, full upstream suite, Windows/macOS test, interoperability test, security approval or recommendation to deploy the entire PR. Symlink cycles, outside-root links, permission failures and time-of-check/time-of-use races were not covered. No maintainer acceptance or merge status is asserted here.

The harness used Node's test runner and ignore@7.0.5 with lifecycle scripts disabled. Temporary test fixtures were removed after execution. No customer credentials were involved.