Repository navigation
v1.2.4
PageSift 1.2.4 Release Notes
Release Version:
v1.2.4
License: AGPL-3.0-only (Commercial licensing available)
Changes since 1.2.3.
Attachment authorization
- Problem:
GET /v2/attachments/{id}checked only the API key and not resource ownership. A valid key for one account could download an attachment produced for another account when the ID was known. IDs are content-derived SHA-256 prefixes and also appear inrewriteMarkdown (asset://<id>), JSON results and logs, so they are not a secret. - Fix: the download endpoint now requires an
(account, attachment)authorization record before reading bytes. A missing or expired record returns the same404response as a missing attachment, with the messageattachment {id} not found or expired, so the response does not reveal whether the attachment exists. Content addressing and cross-account deduplication are unchanged: every account that receives an attachment in a scrape, batch, crawl or parse result gets its own authorization record. - Migration: on startup, missing authorizations are backfilled from saved job results, using the job owner and the attachment IDs in each page. Backfill runs in the background; failures are logged as warnings and do not block startup. Attachments referenced by persisted batch and crawl results keep working for their owning account for the remainder of the 24-hour retention period. Attachments that were only produced by a synchronous scrape are not part of a saved job result and cannot be attributed to an account, so such a link is not backfilled after an upgrade.
- Scope: multi-key deployments are affected. For new results, single-key and no-authentication deployments behave as before: when no API keys are configured, the anonymous identity is used for both writing and checking authorization. MCP attachments remain memory-only and are unaffected.