Actionable comments posted: 1

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In
`@backend/prompt_studio/prompt_studio_core_v2/tests/test_prompt_studio_author.py`:
- Line 104: Update the test setup around the CustomTool and ToolStudioPrompt
timestamp assertions to explicitly backdate the tool via its update operation,
bypassing auto_now and guaranteeing tool.modified_at is earlier than
prompt.modified_at. Keep the strict comparison and endpoint result assertion so
the test verifies the backend selects the prompt timestamp rather than relying
on coincidental row-creation timing.

---

Nitpick comments:
In `@frontend/src/components/widgets/list-view/ListView.jsx`:
- Around line 139-144: Make the metadata Tooltip keyboard-accessible by adding
tabIndex={0} to the InfoCircleOutlined icon used as its child, preserving the
existing tooltip content and icon behavior.

Actionable comments posted: 2

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@frontend/src/components/widgets/list-view/ListView.jsx`:
- Around line 121-126: Update the created-by display expression in ListView to
access the current session email safely when comparing against
item.created_by_email. Use optional chaining on sessionDetails while preserving
the existing “Me”, email, and “-” display behavior.
- Around line 156-184: Update the EditOutlined and ShareAltOutlined onClick
handlers in the list item action rendering to ignore clicks when
item?.is_deprecated is true, while preserving the existing handleEdit and
handleShareClick behavior for active items.

Review — UN-3741 list metadata (no P1s; 3 P2, 3 P3 inline)

Severity: P1 must fix / P2 should fix / P3 optional. Every P2 was independently verified against the code (details in each inline comment).

Checked and cleared:

• Backend annotation/serializer pairing is sound: CustomToolListSerializer is only returned for the list action, and get_queryset always annotates _last_prompt_modified for it; the datetime comparison is tz-aware-safe; a SerializerMethodField shadowing a Meta.fields model field is valid DRF.
• All four ListView consumers (adapters, Prompt Studio, Workflows, Connectors) get both timestamps from their list serializers — no page silently loses the new meta column. The removed CSS classes (list-view-content, adapters-list-title-container, list-item-desc) have no other references.
• moment(...) / formattedDateTime parse DRF's ISO-8601 output cleanly (no DATETIME_FORMAT override, USE_TZ=True) — no "Invalid date" path.
• Prompt reorders / bulk ops do bump modified_at: BaseModelQuerySet patches update()/bulk_update() (utils/models/base_model.py:40-54) and ToolStudioPrompt inherits that manager — so the "honest last modified" holds for those paths too.
• No existing tests were modified or deleted; the new test genuinely exercises the list-action → annotation → serializer path and fails without the fix.
• CodeRabbit's "critical" sessionDetails render-crash is a false positive — ListView.jsx:85 already uses sessionDetails?.email and this PR doesn't touch that line. The deprecated-icon onClick behavior it flagged as major is pre-existing code moved verbatim, out of this PR's scope.

P3 (not inline): title type — this is customer-requested new metadata display, arguably [FEAT] rather than [FIX] per the release-notes standard; author's call since it also corrects the misleadingly stale modified_at.

No blocking issues — zero P1s. The three P2s in my review (silently-ignored ?ordering=modified_at, silent stale fallback in get_modified_at, dead centered prop) are latent/hardening/cosmetic and fine as a follow-up. Backend annotation path, all four ListView consumers, and the new test were verified against the PR head.

Re-approving after re-examining the ordering finding: pre-PR, ?ordering=modified_at only worked for org admins (regular users 500'd on the distinct("tool_id") queryset) and no client sends it — confirmed non-blocking, details in the thread on serializers.py:49. All other findings remain follow-up material (no P1s).

PR Review Toolkit — Code Reviewer, Silent Failure Hunter, Type Design Analyzer, PR Test Analyzer, Comment Analyzer and Code Simplifier. 14 findings below; items already raised by other reviewers are excluded.

The feature is well-motivated and the latest-prompt subquery is the right instinct — the _prompt_count mirroring is clean, the UN-2202 comment in the test documents a genuine trap, and the removed CSS classes were verified unreferenced. Three findings are blocking.

Blocking

1. ListView.jsx:176 — the "Updated" block is presence-keyed, so it silently ships to Workflows and Connectors. Both serializers are fields = "__all__" over BaseModel, so both pages gain the label with no backend change and no test. On Workflows the value is structurally stale in exactly the way this PR fixes for Prompt Studio (content lives in ToolInstance/WorkflowEndpoint, which never write the Workflow row) — a user reconfigures endpoints today and reads "Updated 8 months ago". On Connectors it's bumped by automatic OAuth token refresh (connector_auth_v2/models.py:114), so "Updated 2 minutes ago" appears on connectors nobody touched. Presence of a field isn't evidence of its meaning — make the opt-in explicit.

2. views.py:173 — the derived timestamp travels backwards on prompt delete. ToolStudioPrompt has no save/delete override and no signals, so deleting the newest prompt drops the tool's reported modified_at to the next-newest. The user performs an edit and the row ages by three days.

3. serializers.py:74 — modified_at means two different things on two endpoints. list returns the derived max; every other action goes through CustomToolSerializer (fields = "__all__") and returns the raw row. List then GET makes the timestamp jump backwards, with nothing on the wire to distinguish them.

One fix collapses all three, plus the ?ordering=modified_at breakage already reported: bump the parent row at the source (post_save/post_delete on ToolStudioPrompt → CustomTool.objects.filter(pk=...).update(modified_at=now())). modified_at becomes honest for creates, edits and deletes; both serializers agree because there's only one value; ordering works because it's a real column again; and the annotation, the SerializerMethodField and the view↔serializer coupling all disappear. If you'd rather keep deriving, computing the max in SQL via Greatest/Coalesce restores sortability — but note it does not fix the delete case (#2).

Low — a dead click strip through the middle of every row (ListView.jsx:170); the title doesn't ellipsize despite the comment, so long titles wrap and break row uniformity (ListView.css:59); the adapter fields ship with zero coverage of the list endpoint (adapter_processor_v2/serializers.py:173); the new test's single-tool/single-prompt fixture can't detect a lost OuterRef correlation or a flipped sort (test_prompt_studio_author.py:106).

Nits — dead showOwner guard, mixed presence conventions, eagerly-built tooltip, unguarded "Invalid date" rendering, moment/dayjs split, adapters-list-* prefix on generic classes, and a misleading min-width: 0 comment.

Also worth folding into the centered cleanup already raised: ListView.css:40-46 still targets .ant-list-item-meta, which no longer renders anywhere now that List.Item.Meta is gone — both rules are unreachable. (Outside the diff, so no inline comment.)

Second pass. The parent-bump-at-source redesign is the right call and the backend tests are genuinely good — test_prompt_bump_only_touches_its_own_tool is the one most people omit, and it's the one that would catch a _touch_tool that lost its pk filter. Three findings that clear the bar; the rest of what I checked came back clean (details below).

Verified clean, so nobody re-treads them: reorder does bump (the bulk_update at prompt_studio_v2/helper.py:112 looks like a bypass, but reorder_prompts_helper finishes with prompt_instance.save() at helper.py:64); import_prompts uses objects.create() so the sync ADD path bumps via save(); the deleted CSS classes (.list-view-content, .adapters-list-title-container, .list-item-desc) have zero references left repo-wide; centered is still live and correctly rewired to the row Flex align; AdapterListSerializer leaks nothing (secrets live in adapter_metadata_b, never in that tuple), and the detail serializer already exposed both fields via fields = "__all__", so list is now a strict subset. No deadlock risk on the N-prompt import loop either — I looked for an inverted lock order and there isn't one.

Consider making code comments concise and generic and removing the JIRA dependency if its not truly required

Six-agent PR Review Toolkit pass (code review, silent-failure, type design, tests, comments, simplification) over the backend bump logic and the ListView refactor. No blocking findings — approving.

I verified the load-bearing claims independently rather than trusting them:

• Deletion collector — Django 4.2.30 deletion.py nulls only model._meta.pk.attname (lines 456, 519-521); tool_id_id survives super().delete(), so the delete-bump works. Comment is accurate.
• _base_manager / org scope — CustomToolModelManager mixes in DefaultOrganizationManagerMixin (thread-local scoped); no base_manager_name anywhere in the MRO, so _base_manager is a plain unfiltered Manager. tool_id is a UUID pk. All three sub-claims hold, and test_prompt_bump_survives_missing_org_context genuinely pins it.
• Prompt reorder — flagged by one agent as bypassing the bump via bulk_update; that is wrong. reorder_prompts_helper ends in prompt_instance.save() (prompt_studio_v2/helper.py:64), which fires _touch_tool(). Reorder does bump.
• per_model.get(ToolStudioPrompt._meta.label, 0) — Collector.delete() keys by _meta.label; the 0 default is only hit at zero rows. Correct, and the sync test builds a real CASCADE fixture so the assertion actually discriminates 1 from 2.
• Backdating helper — works only because BaseModelQuerySet.update uses setdefault; had it been an unconditional assign, every > assertion would be vacuous. It is not. Tests are honest.
• CSS — .list-view-content, .adapters-list-title-container, .list-item-desc have zero remaining references; centered is still honored via the new Flex align.

Non-blocking notes, no action needed for this PR:

1. views.py:971 (.update(profile_manager=...) on first-profile creation) bypasses the bump, leaving a stale label on that one path. Pre-existing, display-only, self-healing on the next prompt edit, and the views.py:167-171 comment already warns about this class.
2. Source-bumping is prospective only — existing projects keep their current modified_at until the next prompt write. Worth a thought, but the field never included prompt activity before, so this is strictly an improvement.
3. sync_prompts fires N+2 writes to one CustomTool row per sync. Real write amplification, but prompts-per-tool is in the tens and sync is infrequent — not a throughput concern.
4. timeAgo() is untested and the harness exists (vitest, sibling helpers/requestId.test.js). The strict-parse and moment.min clamp are exactly the non-obvious bits a future cleanup would strip. Cheap to add whenever.

One thing worth fixing outside the diff: the PR description is stale. It still says the Prompt Studio list returns modified_at as max(tool.modified_at, latest prompt.modified_at) via a correlated subquery, and cites a test test_list_modified_at_reflects_prompt_edits that no longer exists. The implementation moved to source-bumping. Worth updating before merge so the description matches what ships.