CommaFeed vs Mattermost

CommaFeed's 7.x line has become a sustained security pass, shipped one fix per patch release. The newest, 7.3.2, closes a cross-user data exposure: users could star or tag entries that were not theirs and then read them, filed as GHSA-prfv-88mm-5gpg. Behind it sit Host header injection on the password recovery endpoint with a new commafeed.password-recovery-public-base-url setting, local address blocking made secure by default alongside Google Reader API support in 7.3.0, and javascript: URL filtering moved to parse time.

The earlier fixes in this run all closed paths where something from outside the instance was trusted too far — feed URLs reaching internal addresses, proxied images, javascript: links, a request header shaping an outbound email. 7.3.2 is a different shape: nothing external is involved, the attacker is a legitimate signed-in user, and the flaw is an ownership check missing on a write path that then leaks through a read path. That is the class of bug you find once you start auditing multi-tenancy rather than input handling, and it suggests the review has moved past the perimeter into the authorization model that the 7.0.0 multi-user rework put in place.

If the audit is now working through ownership checks rather than input validation, the other per-user write paths — subscription and category mutations, saved searches — are the likely next findings. The pattern of shipping each fix as its own patch release with a GHSA reference should continue rather than batching them.