Board view: global session filters (Unread/Running/Pinned/Recent) have no visible effect

[vidanov]

Summary

In the board view (Trello-style tag-column layout), toggling the global session filters — Unread, In progress, Pinned, or Recent — from the sort/filter dropdown does not visually change which sessions are shown in the columns.

Steps to reproduce

1. Create at least two tag columns (board view activates when orderedColumns.length > 0).
2. Have a mix of sessions: some unread, some running, some pinned.
3. Open the Filter dropdown (ListFilter icon, top-right of the sidebar search bar).
4. Toggle any filter — e.g. Unread or Pinned.
5. Expected: columns show only sessions matching the active filter.
6. Actual: all sessions remain visible; the filter has no apparent effect.

Root cause analysis

The filter predicate lives in filteredSlots (ChatSidebar.tsx ~L1215):

const activeFilterDefs = SESSION_FILTERS.filter(filterDef => activeFilters.has(filterDef.key))
return enrichedSlots.filter(slot => {
  if (activeFilterDefs.length > 0 && !activeFilterDefs.some(filterDef => slot[filterDef.key])) return false
  ...
})

Board columns then derive their slot list from filteredSlots:

// ChatSidebar.tsx ~L2813
const colSlots = filteredSlots.filter(s => columnMatches(col, s.tags || []))

This chain is architecturally correct. The likely causes of the observed no-op are:

1. Column with tag_ids: [] matches all slots unconditionally

columnMatches (L1184) returns true immediately when a column has no tag filter configured:

if (!col.tag_ids || col.tag_ids.length === 0) return true

A new or unconfigured column ignores all global filters because it passes every slot through. Users who have columns without tag filters will see all sessions regardless of global filter state.

2. Silent tag-ID drop in _normalize_column (backend)

When updating a column via PATCH, the backend validates tag_ids against current known tag IDs:

# chat_tags.py L183
cleaned["tag_ids"] = [str(t) for t in tag_ids if isinstance(t, str) and t in valid_ids]

Any ID not present in state._tags is silently dropped. If tags are deleted while a column still references them, the column's filter rules become empty ([]), making it show all sessions again — with no feedback to the user.

Suggested fixes

Frontend: surface "no filter configured" state on a column

When col.tag_ids.length === 0 and !col.include_untagged, show a visual cue in the column header (e.g. a dimmed "All sessions" label) so users know that column is unfiltered and global filters won't narrow it further.

Backend: warn on dropped tag IDs

In _normalize_column, detect when tag IDs were provided but all dropped (unknown IDs), and either return a 400 or include a warnings field in the response so the client can surface it.

Consideration: should global filters apply even to columns with tag_ids: []?

An open design question: should a global "Pinned" filter hide non-pinned sessions even in an "All sessions" column? Currently filteredSlots applies before columnMatches, so global filters DO reduce the set passed to every column. If the filter appears not to work, the column itself may be the issue (see point 1 above).

Environment

• KiroCrew version: 0.1.0-insider.3 (also reproduced on 0.1.0.dev20260728)
• Tested in: dashboard board view with 2+ tag columns

[NicholasRBowers]

🤖 Kiro Crew Auto-Pipeline [operator: NicholasRBowers]: Routing to needs-investigation — the report is good, but its leading hypothesis does not survive a read of the code, and the next person should not spend their time there.

Ruled out: cause 1 (columnMatches returning true for a column with no tag_ids) cannot produce this symptom. Board columns derive from the ALREADY-filtered set — ChatSidebar.tsx:3538 is filteredSlots.filter(s => columnMatches(col, s.tags || [])), and the global-filter predicate is applied upstream at ChatSidebar.tsx:1524. So an unconfigured column shows every filtered slot, not every slot; passing all slots through the tag test cannot re-admit a session the global filter already removed. Your own note that the chain is architecturally correct is right, and it is precisely why that candidate fails.

Worth noting what would be observable if the predicate itself were broken: ChatSidebar.tsx:1087 computes the filter chip counts with the SAME slot[filterDef.key] access, so if those boolean properties were missing from the enriched slots, every filter chip would read 0 — visible, and not something you reported. That makes a broken key access unlikely too.

What is left needs runtime evidence rather than static reading, which is why this is not going down the automated-fix path: whether toggling actually updates the active-filter state in the board branch, and whether some other board render path reads an unfiltered array. A useful first datapoint: with a filter on, do the chip counts change and does the FLAT (non-board) lane filter correctly? If flat filters and board does not, the divergence is in the board branch; if neither filters, it is in the toggle-to-state wiring.

[CrysisDeu]

🤖 Kiro Crew Auto-Pipeline [operator: CrysisDeu#0c98c3a2]: Verified fixed on current main. Board columns now derive their members from the globally-filtered list: website/src/pages/ChatSidebar.tsx:5210 computes const colSlots = filteredSlots.filter(s => columnMatches(col, s)), and filteredSlots (~:2442) applies every declared filter dimension including the Unread/In progress/Pinned/Recent predicates (filtersRow with activeFilterDefs, ~:2366-2396). Toggling a global session filter therefore narrows board columns as expected. Closing as fixed; please reopen if you can still reproduce on a current build.