- CLAUDE.md: document planned user-scoped token scope (account_id filter) - TODO.md: add #34 spec, drop stale SUGGESTIONS.md reference - SUGGESTIONS.md: deleted — fully superseded by TODO.md + CLAUDE.md Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4.1 KiB
TODO — Pending features and sustainability
Quick overview of what's still to be done.
Recently completed
Memory exhaustion during large M365 scans ✅
Six root causes fixed in scan_engine.py and document_scanner.py:
- Email body HTML stripped at collection time (
bodykey deleted from each message dict before it enterswork_items; plain text stored as_precomputed_bodyinstead) work_itemslist converted to adequebefore processing so each item is released immediately afterpopleft()del contentadded in file-processing branch as soon as raw bytes are no longer needed (before NER/PII counting)del body_textadded after email body is fully consumed- PDF OCR page images (
PIL.Image) nulled out one by one after OCR instead of holding all pages in RAM - Memory guard using
psutilskips file downloads when < 300 MB RAM is available
Still open: The collection phase itself is still a "gather all, then process" loop. For very large tenants (>500k emails) the pre-extracted plain text in work_items could still be significant. The complete fix is to process each user's emails/files inline as they are fetched (generator/streaming pattern) rather than accumulating them into work_items first — estimated 1–2 days of refactor.
Pending
#15 — Scan profiles ✅
Named, reusable scan configurations. Full spec in SUGGESTIONS.md §15.
Size: Large · Priority: High
#23 — Google Workspace role classification + cross-platform identity mapping ✅
Full spec in SUGGESTIONS.md §23.
Size: Large · Priority: Medium
#27 — Migrate i18n format from .lang to JSON ✅
Full spec in SUGGESTIONS.md §27.
Size: Medium · Priority: Low
#29 — Rename skus/ → classification/ ✅
Full spec in SUGGESTIONS.md §29.
Size: Small · Priority: Low
#33 — Read-only viewer mode with PIN/token URL ✅
A shareable URL (token-protected) or numeric PIN that gives a DPO, school principal, or compliance coordinator read-only access to the results grid — with disposition tagging but without scan controls, credentials, or delete access. Full spec in SUGGESTIONS.md §33.
Size: Medium · Priority: Medium
OneDrive 404 errors — investigate and handle appropriately ✅
404 on drive/root/delta during delta scans was being broadcast as a red scan_error. Root cause: _get() hit raise_for_status() for 404s, which fell through to the generic except Exception handler in _scan_user_onedrive. The full-scan path silently swallowed the same 404 via except Exception: return in _iter_drive_folder_for.
Fixed by adding M365DriveNotFound(M365Error) exception, raising it from _get() on 404, and catching it explicitly in _scan_user_onedrive with a lower-severity scan_phase broadcast ("OneDrive (user): not provisioned — skipped") instead of a red error card.
#34 — User-scoped viewer tokens
Extend viewer token scope from {"role": "student"|"staff"} to also support {"user": "alice@school.dk"}, filtering flagged_items by account_id. Lets a single employee see only their own flagged files.
Infrastructure already in place: account_id is an indexed column on flagged_items, populated for M365 (UPN) and Google (email). File-scan items have account_id = "" and won't appear in user-scoped views — document this in the token-creation UI.
Changes needed:
- Token creation UI — add a "specific user" option (email input) alongside the role dropdown
GET /api/db/flagged— filter byaccount_idwhensession["viewer_scope"].get("user")is set (same pattern as existing role filter)- Viewer header — show locked identity (similar to locked
#filterRolefor role-scoped tokens)
Size: Small · Priority: Medium
#32 — Windowed mode for Profiles, Sources, and Settings ✗ Won't do
The workflow is sequential (configure → scan → review), not parallel — there is no realistic scenario where a modal and the results grid need to be open simultaneously. The Sources panel is already visible in the sidebar. Option A (the least-work path) still loads the full 3800-line JS stack twice. Closed.