Repository navigation
Conversation
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Check the description: it says a line leaves [DESIGN-REVIEWED] 6327261 Description read[DESCRIPTION-READ] 1125d879bcbeee95c4c0f1d0421fae136d572e511e69a39d2f912ac45880cf10 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
UX Review (Opus 5.5) — 🟡 CONCERNSUX-level review of UX-Verdict: CONCERNS The unpinned-model row's "(agent default)" made no sense to the blind reader and seems to clash with the footer's "Default". The model rows were read correctly but only as "a guess". Watch
[UX-REVIEWED] 6327261 Description read[DESCRIPTION-READ] f67f32a1cf4e52f89ca24ea87df85b3769ab010b5759ecb5fb0c465044354196 That sha256 names everything this verdict read: the PR description, plus the 10 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS The description says a line is removed from What this change shipsInventory (8 items) — 8 justifiedIntent: find the sessions that run on a given model, or on any model but that one, from the sidebar filter menu (#16527). ADDITION.
[FIRST-PRINCIPLES-REVIEWED] 6327261 Description read[DESCRIPTION-READ] 3204d9ce7d789bc3db2a62ddf07e1ca3eb739d5d3f2d6dc74e6a71d0f12494dc That sha256 names the PR description this verdict read. Recompute it with |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
57ea979 to
51fe3ca
Compare
|
|
|
|
|
|
a7dad6e to
2b85d4a
Compare
|
|
|
|
2b85d4a to
54fb8b0
Compare
|
|
|
|
|
|
54fb8b0 to
c28d604
Compare
4bcff13 to
1cc4dc3
Compare
1cc4dc3 to
6458c44
Compare
6458c44 to
57136f7
Compare
The sidebar filters by status, folder and tag, not by model. Someone who runs two models and wants the sessions still pinned to an older one has to open each session to find them. The filter menu gains a Models section listing only the models some session uses, most used first with a session count each, so it doubles as a read of what the sessions run on. Rows are a union, like tags. A "Hide selected models" row, offered once a row is ticked, inverts the selection, so "every session not on model X" is one tick plus the toggle. The selection and its direction persist under one localStorage key, and a selected model no session uses any more is ignored rather than allowed to blank the list. One aggregate chip names the selection, or its negation, and clears it. Design: - The vocabulary is the slot list, not the registry: a registry row for a model no session uses could only blank the list. - Slots group by `normalizeModelKey`, so an alias and its provider id fold onto one row; an unpinned slot and an explicit `auto` share the Default row, since the filter asks which model runs a session. - A model the registry knows is labelled with its display name through a new `displayName` reader; an unknown id keeps its id, minus a routing prefix. - The filter is a fifth `filterDimensions` entry, so narrowing, hiding and reveal-clear derive from the one declaration.
57136f7 to
6327261
Compare
|
Rebased
Three conflicts came up. Each was main changing text next to this PR's additions, and I kept both sides:
Checks run on the rebased tree: vitest If you had local work on this branch, run |
|
Taking this over in #19605 since it has been idle; you are credited as co-author on the commits. Thanks for the groundwork. |
|
Thanks for this, @smeyffret! The model filter for the sidebar landed in #19605 (commit 76f947d), which closed #16527, so we're closing this draft as superseded. If something here isn't covered by #19605, feel free to reopen. |
Problem / Motivation
Goal: Find the sessions that run on a given model, or on any model but that one, from the sidebar filter menu.
The sidebar filters by status, folder and tag, not by model. Someone who runs two models and wants the sessions still pinned to an older one has to open each session to read its model. There is also no place that shows which models the sessions use at all.
Why it matters
Moving sessions off a retired model is a per-session action, and finding them is the slow part. A model filter turns the hunt into one tick. The same list, with a count per model, answers "what are my sessions running on" without opening any of them.
Not a goal
What changed (motivation → approach → change)
mc-session-model-filterThe menu lists only models in use, never the registry catalog. A catalog row for a model no session runs could only blank the list. The rows are built from the slot list on the client, so the section doubles as a read of what is in use. The
models.tsowner groups slots bynormalizeModelKey, so an alias and its provider id share one row, and an unpinned slot and an explicitautoshare the No model chosen row. The key is the stored pin and only that. Two things the composer folds intoautostay apart here on purpose: a pin the account can no longer run (model_withheld: true, a retired or unentitled model) keeps its own row, because such a session is exactly the one this filter exists to find; and the model a pinless session is served (served_model) is not used, because a session that follows the default moves by itself. Both fields also exist only while a session is live, so reading them would make the counts shift with runtime state rather than with anything the person did. A model row is labelled with the id the session carries, minus a routing prefix: the composer footer and the assistant header label a model by id, so the filter names it the same way. The row for unpinned sessions is "No model chosen (agent default)": neither "Default" alone (the composer footer's word for the resolved model) nor "pinned" (the sidebar's word for pinned sessions) is free, and the parenthesis says what such a session runs on.Several rows are a union, like tags. "Hide selected models" inverts it, so "every session not on model X" is one tick plus the toggle. The direction lives in the same stored object as the selection: deselecting the last model drops it, a stored key no session uses any more is ignored and cannot keep "Not X" alive with nothing selected (the invert row then reads unticked and the tooltips say "show only"), and a stored direction with no keys reads as off. A session from a peer gateway carries no model in its row (
_clean_peer_slotallowlists none), so it is hidden while the filter narrows, in either direction, rather than counted as unpinned.The filter is a fifth entry in
ChatSidebar.tsx'sfilterDimensions, so narrowing, hiding and reveal-clear come from the one declaration.filters.tsxholds the state,persistence.tsthe reader and writer. The catalogs underwebsite/src/i18n/locales/gain eight keys in all twelve locales. The feature map and the sidebar owner table inhistory.mdname the new section and module.One line leaves
.github/coverage-baselines/backend.txt: the Coverage Gate on this PR's run reportedsrc/kiro_crew/apps/hooks_integration.pyat 82.3%, above the 80% floor, and the gate fails until a baselined file that now meets the floor is removed. The entry was pruned with the gate's own--update-baselineprocedure against this run'scoverage-backendartifact. No Python changes here; the file crossed the floor onmain.Backwards compatibility
Compatible: no stored key, API or component contract changes; the new localStorage key is read with the same fall-back-to-no-filter rule as the tag filter.
Removes nothing: every change adds a menu section, a chip and a filter dimension.
Tests
ChatSidebar.modelFilter.test.tsx(14 tests) locks in: rows list only models in use, most used first, with spelling variants folded onto one row; a tick narrows the list; several ticks are a union; the invert row is disabled until a tick, then inverts the list, marking the ticked rows as hidden; a withheld pin stays under its own model; deselecting the last model drops the direction; selection and direction survive a remount; a stored key no session uses is ignored, in include and in exclude mode; a stored direction with no keys reads as off; the chip names the selection or its negation and clears it in one control; the tooltip names what a tick does in exclude mode; a reveal of a hidden session clears the filter; a peer-gateway row never passes a narrowing filter and adds no count to the vocabulary.ChatSidebar.filterDimensions.test.tsxpins five dimensions;ChatSidebar.ownerComposition.test.tspinsuseSidebarModelsin the hook order.Manual verification
Screenshots below come from the built SPA served against fixture slots (six sessions on three models plus one unpinned), light and dark, at 1400x900.
Screenshots / video
Filter menu with the Filter by model section:
One model ticked; the Hide selected models row becomes enabled:
List narrowed to that model, with the chip:
Hide selected models switched on; the row heads the section with the same checkbox square as the model rows, so the mode is read before the ticks, and the ticked model reads as hidden:
Inverted: every session not on that model:
Related Issues
Closes #16527
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)