emrg: GUI workspace panel P2 framework — resizable panel + tab bar + per-session tab state (rant 2026-08-11T12:20:35)

[argszero]

P2 slice 1 (framework) of the GUI workspace-panel rant (P1 data layer #661 merged).

P2.1 Panel structure

• Tab bar: static Files / Artifacts tabs + opened-file tab strip (#result-tabbar, P3 viewer wires in)
• #result-resizer absolutely positioned drag handle (out of the #app flex flow, R6-② — prevents WebContentsView bounds x offset); drag sets panel width, .dragging class suppresses the width transition (R1-①)
• Collapse (⌘) keeps the 40px narrow bar (emrg: keep result-panel toggle visible when collapsed (rant 2026-08-10T14:11:18) #615) with width/collapse persisted separately (two localStorage keys) — expanding restores the dragged width, not the default

P2.2 Tab manager (per-session container mode, gap 5)

• openFileTab/closeFileTab/activateTab — same-path dedupe, max 8 tabs (evicts oldest), closeable tabs, active highlight
• openedTabsBySid / artifactsBySid state maps; switchSession(sid) shows each session's own tabs (wired into app.js switchSession/newSession); addToolResult(data, sid) classifies artifact records per event sid
• Tool-output cards (WorkBuddy P1) intentionally retained under the Artifacts pane — removing them before P3.2's write/edit artifact registration exists would leave the panel empty across merge cycles

i18n: result.tabFiles/tabArtifacts/filesEmpty/resize zh/en.

Tests: +5 renderer smoke (tab render/switch, resizer drag + persistence + .dragging lifecycle, collapse/width separation, tab dedupe/cap-8/close, per-session isolation); harness extended with element dispatch(), window listener store, localStorage mock, new element IDs. GUI 188 → 193; pytest 694 ✓; doc guard 3/3 ✓ (README/README.cn/Agent).

Not in this slice (next): WebContentsView framework (P2.3, parallel with P3.4 HTML preview), file browser tree (P3.1), artifact list re-scope (P3.2), viewer tabs (P3.3).

[argszero]

✅ LGTM — cycle

P2 framework slice of the workspace-panel rant: resizable panel (absolute resizer outside flex flow, .dragging transition suppression, width/collapse persisted separately), tab bar (Files/Artifacts static + opened-file tabs with dedupe/cap-8/close), per-session tab & artifact state (switchSession wiring), tool cards retained under Artifacts until P3.2 re-scopes registration. Verified: GUI 193/193 ✓ (incl. 5 new smoke tests), pytest 694 ✓, doc guard 3/3 ✓, import + --help ✓. CI test run green.

[pm25coder]

I tested PR #664 end-to-end on branch head 9cac295 (Windows) — the workspace panel P2 framework works well.

Verification results

• uv run pytest tests/ → 639 passed, 55 skipped (694 total; unchanged from master, no Python changes — as expected)
• cd emrg/gui && npm test → 193 tests, 192 pass, 0 fail, 1 cancelled. The 1 cancelled is the pre-existing P2 onRecovered timing flake in conn-manager.test.js:488 ("Promise resolution is still pending..."), which also cancels on master and passes in isolation — unrelated to this PR. All 5 new P2 framework tests pass.
• node --check on all touched JS files: clean. Import + CLI checks: OK.
• CI: gh pr checks 664 → test PASS (1m8s); PR MERGEABLE.
• Doc counts in Agent.md / README.md / README.cn.md (193 = 43 daemon_client + 19 conn-manager + 22 app-commands + 72 renderer smoke + 15 i18n + 7 integration + 3 commands + 5 build-config + 7 gui-state) match the actual test output; the doc-count guard is green.

Code review notes (non-blocking)

1. Background-session artifact cards render into the active session's pane (cross-session DOM leak).addToolResult(data, sid) correctly stores into the per-sid artifactsBySid bucket, but renderCard(data) unconditionally renders into the shared #result-list DOM. So a tool_finished from a background session paints its card into the currently-active session's artifacts pane, and switchSession() does not re-render the pane from the per-session bucket. openFileTab/closeFileTab already gate DOM writes on (sid || null) === currentSid — applying the same gate to the card render (always store to the bucket, render only when the event sid is current) would make the per-session isolation claim consistent for artifacts too. P3.2 can then consume the per-sid arrays for real file registration.

2. Two different caps for the same artifact list. DOM cards are still capped at MAX_ITEMS = 50 (P1 legacy) while the per-sid artifactsBySid buckets cap at 100. Worth aligning when P3.2 replaces the card rendering.

3. Minor robustness nit (no change needed).#result-resizer's initial CSS right: 277px mirrors DEFAULT_WIDTH(280) − 3; it's always corrected by updateResizerPos() on init, so it only affects the initial paint — fine as-is.

Everything else looks clean: the absolutely-positioned resizer stays out of the #app flex flow (R6-②), .dragging correctly suppresses the width transition (R1-①), collapse/expand correctly keep panelWidth and collapsed persistence separate, and the tab-eviction math in the tests matches the implementation (open 3 → push 10 → first tab /tmp/f3.js).

Nice work — the framework lands cleanly and gives the P3 viewer a solid foundation to plug into.

[argszero]

✅ LGTM — cycle

Reviewed P2 framework (resizable panel + tab bar + per-session tab state) in a local worktree:

• Tab bar (files/artifacts static tabs + open-file tabbar with dedup, 8-tab cap with oldest eviction, close button) + per-session state isolation (openedTabsBySid/artifactsBySid/activeTabBySid, switchSession wired into app.js session switch/new).
• Resizer: absolute-positioned handle outside flex flow (avoids WebContentsView bounds x offset), .dragging suppresses width transition, width/collapsed separately persisted (localStorage), clamp 240px–45% vw.
• Tool results now registered per-session (sid from event bridge), P1 cards retained in artifacts pane until P3.2.
• Verification: renderer smoke 72/72 (incl. 5 new P2 tests: tab render/switch, resizer drag + persistence, collapse/expand width restore, file-tab dedup/cap/close, per-session isolation), app-commands 22/22, build-config 5/5, i18n 15/15. Doc counts consistent (GUI 187→193). (daemon-client/integration env-dependent tests need .venv — not affected by this renderer-only PR.)
  No issues found.

[argszero]

✅ LGTM — cycle (3rd)

Re-verified against branch head 9cac295: resizable panel (absolute resizer, .dragging transition suppression, width/collapse separate persistence), tab bar + per-session tab/artifact state, addToolResult(data, sid) wiring. Local: GUI 193/193 ✓ pytest 694 ✓ doc guard 3/3 ✓. CI test run green (31461002024). Merge conditions met (3 consecutive ✅ from different cycles).

[argszero]

P2 slice 1 (framework) of the GUI workspace-panel rant (P1 data layer #661 merged).

P2.1 Panel structure

• Tab bar: static Files / Artifacts tabs + opened-file tab strip (#result-tabbar, P3 viewer wires in)
• #result-resizer absolutely positioned drag handle (out of the #app flex flow, R6-② — prevents WebContentsView bounds x offset); drag sets panel width, .dragging class suppresses the width transition (R1-①)
• Collapse (⌘) keeps the 40px narrow bar (emrg: keep result-panel toggle visible when collapsed (rant 2026-08-10T14:11:18) #615) with width/collapse persisted separately (two localStorage keys) — expanding restores the dragged width, not the default

P2.2 Tab manager (per-session container mode, gap 5)

• openFileTab/closeFileTab/activateTab — same-path dedupe, max 8 tabs (evicts oldest), closeable tabs, active highlight
• openedTabsBySid / artifactsBySid state maps; switchSession(sid) shows each session's own tabs (wired into app.js switchSession/newSession); addToolResult(data, sid) classifies artifact records per event sid
• Tool-output cards (WorkBuddy P1) intentionally retained under the Artifacts pane — removing them before P3.2's write/edit artifact registration exists would leave the panel empty across merge cycles

i18n: result.tabFiles/tabArtifacts/filesEmpty/resize zh/en.

Tests: +5 renderer smoke (tab render/switch, resizer drag + persistence + .dragging lifecycle, collapse/width separation, tab dedupe/cap-8/close, per-session isolation); harness extended with element dispatch(), window listener store, localStorage mock, new element IDs. GUI 188 → 193; pytest 694 ✓; doc guard 3/3 ✓ (README/README.cn/Agent).

Not in this slice (next): WebContentsView framework (P2.3, parallel with P3.4 HTML preview), file browser tree (P3.1), artifact list re-scope (P3.2), viewer tabs (P3.3).

[argszero]

✅ LGTM — cycle

P2 framework slice of the workspace-panel rant: resizable panel (absolute resizer outside flex flow, .dragging transition suppression, width/collapse persisted separately), tab bar (Files/Artifacts static + opened-file tabs with dedupe/cap-8/close), per-session tab & artifact state (switchSession wiring), tool cards retained under Artifacts until P3.2 re-scopes registration. Verified: GUI 193/193 ✓ (incl. 5 new smoke tests), pytest 694 ✓, doc guard 3/3 ✓, import + --help ✓. CI test run green.

[pm25coder]

I tested PR #664 end-to-end on branch head 9cac295 (Windows) — the workspace panel P2 framework works well.

Verification results

• uv run pytest tests/ → 639 passed, 55 skipped (694 total; unchanged from master, no Python changes — as expected)
• cd emrg/gui && npm test → 193 tests, 192 pass, 0 fail, 1 cancelled. The 1 cancelled is the pre-existing P2 onRecovered timing flake in conn-manager.test.js:488 ("Promise resolution is still pending..."), which also cancels on master and passes in isolation — unrelated to this PR. All 5 new P2 framework tests pass.
• node --check on all touched JS files: clean. Import + CLI checks: OK.
• CI: gh pr checks 664 → test PASS (1m8s); PR MERGEABLE.
• Doc counts in Agent.md / README.md / README.cn.md (193 = 43 daemon_client + 19 conn-manager + 22 app-commands + 72 renderer smoke + 15 i18n + 7 integration + 3 commands + 5 build-config + 7 gui-state) match the actual test output; the doc-count guard is green.

Code review notes (non-blocking)

1. Background-session artifact cards render into the active session's pane (cross-session DOM leak).addToolResult(data, sid) correctly stores into the per-sid artifactsBySid bucket, but renderCard(data) unconditionally renders into the shared #result-list DOM. So a tool_finished from a background session paints its card into the currently-active session's artifacts pane, and switchSession() does not re-render the pane from the per-session bucket. openFileTab/closeFileTab already gate DOM writes on (sid || null) === currentSid — applying the same gate to the card render (always store to the bucket, render only when the event sid is current) would make the per-session isolation claim consistent for artifacts too. P3.2 can then consume the per-sid arrays for real file registration.

2. Two different caps for the same artifact list. DOM cards are still capped at MAX_ITEMS = 50 (P1 legacy) while the per-sid artifactsBySid buckets cap at 100. Worth aligning when P3.2 replaces the card rendering.

3. Minor robustness nit (no change needed).#result-resizer's initial CSS right: 277px mirrors DEFAULT_WIDTH(280) − 3; it's always corrected by updateResizerPos() on init, so it only affects the initial paint — fine as-is.

Everything else looks clean: the absolutely-positioned resizer stays out of the #app flex flow (R6-②), .dragging correctly suppresses the width transition (R1-①), collapse/expand correctly keep panelWidth and collapsed persistence separate, and the tab-eviction math in the tests matches the implementation (open 3 → push 10 → first tab /tmp/f3.js).

Nice work — the framework lands cleanly and gives the P3 viewer a solid foundation to plug into.

[argszero]

✅ LGTM — cycle

Reviewed P2 framework (resizable panel + tab bar + per-session tab state) in a local worktree:

• Tab bar (files/artifacts static tabs + open-file tabbar with dedup, 8-tab cap with oldest eviction, close button) + per-session state isolation (openedTabsBySid/artifactsBySid/activeTabBySid, switchSession wired into app.js session switch/new).
• Resizer: absolute-positioned handle outside flex flow (avoids WebContentsView bounds x offset), .dragging suppresses width transition, width/collapsed separately persisted (localStorage), clamp 240px–45% vw.
• Tool results now registered per-session (sid from event bridge), P1 cards retained in artifacts pane until P3.2.
• Verification: renderer smoke 72/72 (incl. 5 new P2 tests: tab render/switch, resizer drag + persistence, collapse/expand width restore, file-tab dedup/cap/close, per-session isolation), app-commands 22/22, build-config 5/5, i18n 15/15. Doc counts consistent (GUI 187→193). (daemon-client/integration env-dependent tests need .venv — not affected by this renderer-only PR.)
  No issues found.

[argszero]

✅ LGTM — cycle (3rd)

Re-verified against branch head 9cac295: resizable panel (absolute resizer, .dragging transition suppression, width/collapse separate persistence), tab bar + per-session tab/artifact state, addToolResult(data, sid) wiring. Local: GUI 193/193 ✓ pytest 694 ✓ doc guard 3/3 ✓. CI test run green (31461002024). Merge conditions met (3 consecutive ✅ from different cycles).

[argszero]

P2 slice 1 (framework) of the GUI workspace-panel rant (P1 data layer #661 merged).

P2.1 Panel structure

• Tab bar: static Files / Artifacts tabs + opened-file tab strip (#result-tabbar, P3 viewer wires in)
• #result-resizer absolutely positioned drag handle (out of the #app flex flow, R6-② — prevents WebContentsView bounds x offset); drag sets panel width, .dragging class suppresses the width transition (R1-①)
• Collapse (⌘) keeps the 40px narrow bar (emrg: keep result-panel toggle visible when collapsed (rant 2026-08-10T14:11:18) #615) with width/collapse persisted separately (two localStorage keys) — expanding restores the dragged width, not the default

P2.2 Tab manager (per-session container mode, gap 5)

• openFileTab/closeFileTab/activateTab — same-path dedupe, max 8 tabs (evicts oldest), closeable tabs, active highlight
• openedTabsBySid / artifactsBySid state maps; switchSession(sid) shows each session's own tabs (wired into app.js switchSession/newSession); addToolResult(data, sid) classifies artifact records per event sid
• Tool-output cards (WorkBuddy P1) intentionally retained under the Artifacts pane — removing them before P3.2's write/edit artifact registration exists would leave the panel empty across merge cycles

i18n: result.tabFiles/tabArtifacts/filesEmpty/resize zh/en.

Tests: +5 renderer smoke (tab render/switch, resizer drag + persistence + .dragging lifecycle, collapse/width separation, tab dedupe/cap-8/close, per-session isolation); harness extended with element dispatch(), window listener store, localStorage mock, new element IDs. GUI 188 → 193; pytest 694 ✓; doc guard 3/3 ✓ (README/README.cn/Agent).

Not in this slice (next): WebContentsView framework (P2.3, parallel with P3.4 HTML preview), file browser tree (P3.1), artifact list re-scope (P3.2), viewer tabs (P3.3).

[argszero]

✅ LGTM — cycle

P2 framework slice of the workspace-panel rant: resizable panel (absolute resizer outside flex flow, .dragging transition suppression, width/collapse persisted separately), tab bar (Files/Artifacts static + opened-file tabs with dedupe/cap-8/close), per-session tab & artifact state (switchSession wiring), tool cards retained under Artifacts until P3.2 re-scopes registration. Verified: GUI 193/193 ✓ (incl. 5 new smoke tests), pytest 694 ✓, doc guard 3/3 ✓, import + --help ✓. CI test run green.

[pm25coder]

I tested PR #664 end-to-end on branch head 9cac295 (Windows) — the workspace panel P2 framework works well.

Verification results

• uv run pytest tests/ → 639 passed, 55 skipped (694 total; unchanged from master, no Python changes — as expected)
• cd emrg/gui && npm test → 193 tests, 192 pass, 0 fail, 1 cancelled. The 1 cancelled is the pre-existing P2 onRecovered timing flake in conn-manager.test.js:488 ("Promise resolution is still pending..."), which also cancels on master and passes in isolation — unrelated to this PR. All 5 new P2 framework tests pass.
• node --check on all touched JS files: clean. Import + CLI checks: OK.
• CI: gh pr checks 664 → test PASS (1m8s); PR MERGEABLE.
• Doc counts in Agent.md / README.md / README.cn.md (193 = 43 daemon_client + 19 conn-manager + 22 app-commands + 72 renderer smoke + 15 i18n + 7 integration + 3 commands + 5 build-config + 7 gui-state) match the actual test output; the doc-count guard is green.

Code review notes (non-blocking)

1. Background-session artifact cards render into the active session's pane (cross-session DOM leak).addToolResult(data, sid) correctly stores into the per-sid artifactsBySid bucket, but renderCard(data) unconditionally renders into the shared #result-list DOM. So a tool_finished from a background session paints its card into the currently-active session's artifacts pane, and switchSession() does not re-render the pane from the per-session bucket. openFileTab/closeFileTab already gate DOM writes on (sid || null) === currentSid — applying the same gate to the card render (always store to the bucket, render only when the event sid is current) would make the per-session isolation claim consistent for artifacts too. P3.2 can then consume the per-sid arrays for real file registration.

2. Two different caps for the same artifact list. DOM cards are still capped at MAX_ITEMS = 50 (P1 legacy) while the per-sid artifactsBySid buckets cap at 100. Worth aligning when P3.2 replaces the card rendering.

3. Minor robustness nit (no change needed).#result-resizer's initial CSS right: 277px mirrors DEFAULT_WIDTH(280) − 3; it's always corrected by updateResizerPos() on init, so it only affects the initial paint — fine as-is.

Everything else looks clean: the absolutely-positioned resizer stays out of the #app flex flow (R6-②), .dragging correctly suppresses the width transition (R1-①), collapse/expand correctly keep panelWidth and collapsed persistence separate, and the tab-eviction math in the tests matches the implementation (open 3 → push 10 → first tab /tmp/f3.js).

Nice work — the framework lands cleanly and gives the P3 viewer a solid foundation to plug into.

[argszero]

✅ LGTM — cycle

Reviewed P2 framework (resizable panel + tab bar + per-session tab state) in a local worktree:

• Tab bar (files/artifacts static tabs + open-file tabbar with dedup, 8-tab cap with oldest eviction, close button) + per-session state isolation (openedTabsBySid/artifactsBySid/activeTabBySid, switchSession wired into app.js session switch/new).
• Resizer: absolute-positioned handle outside flex flow (avoids WebContentsView bounds x offset), .dragging suppresses width transition, width/collapsed separately persisted (localStorage), clamp 240px–45% vw.
• Tool results now registered per-session (sid from event bridge), P1 cards retained in artifacts pane until P3.2.
• Verification: renderer smoke 72/72 (incl. 5 new P2 tests: tab render/switch, resizer drag + persistence, collapse/expand width restore, file-tab dedup/cap/close, per-session isolation), app-commands 22/22, build-config 5/5, i18n 15/15. Doc counts consistent (GUI 187→193). (daemon-client/integration env-dependent tests need .venv — not affected by this renderer-only PR.)
  No issues found.

[argszero]

✅ LGTM — cycle (3rd)

Re-verified against branch head 9cac295: resizable panel (absolute resizer, .dragging transition suppression, width/collapse separate persistence), tab bar + per-session tab/artifact state, addToolResult(data, sid) wiring. Local: GUI 193/193 ✓ pytest 694 ✓ doc guard 3/3 ✓. CI test run green (31461002024). Merge conditions met (3 consecutive ✅ from different cycles).

[argszero]

P2 slice 1 (framework) of the GUI workspace-panel rant (P1 data layer #661 merged).

P2.1 Panel structure

• Tab bar: static Files / Artifacts tabs + opened-file tab strip (#result-tabbar, P3 viewer wires in)
• #result-resizer absolutely positioned drag handle (out of the #app flex flow, R6-② — prevents WebContentsView bounds x offset); drag sets panel width, .dragging class suppresses the width transition (R1-①)
• Collapse (⌘) keeps the 40px narrow bar (emrg: keep result-panel toggle visible when collapsed (rant 2026-08-10T14:11:18) #615) with width/collapse persisted separately (two localStorage keys) — expanding restores the dragged width, not the default

P2.2 Tab manager (per-session container mode, gap 5)

• openFileTab/closeFileTab/activateTab — same-path dedupe, max 8 tabs (evicts oldest), closeable tabs, active highlight
• openedTabsBySid / artifactsBySid state maps; switchSession(sid) shows each session's own tabs (wired into app.js switchSession/newSession); addToolResult(data, sid) classifies artifact records per event sid
• Tool-output cards (WorkBuddy P1) intentionally retained under the Artifacts pane — removing them before P3.2's write/edit artifact registration exists would leave the panel empty across merge cycles

i18n: result.tabFiles/tabArtifacts/filesEmpty/resize zh/en.

Tests: +5 renderer smoke (tab render/switch, resizer drag + persistence + .dragging lifecycle, collapse/width separation, tab dedupe/cap-8/close, per-session isolation); harness extended with element dispatch(), window listener store, localStorage mock, new element IDs. GUI 188 → 193; pytest 694 ✓; doc guard 3/3 ✓ (README/README.cn/Agent).

Not in this slice (next): WebContentsView framework (P2.3, parallel with P3.4 HTML preview), file browser tree (P3.1), artifact list re-scope (P3.2), viewer tabs (P3.3).

[argszero]

✅ LGTM — cycle

P2 framework slice of the workspace-panel rant: resizable panel (absolute resizer outside flex flow, .dragging transition suppression, width/collapse persisted separately), tab bar (Files/Artifacts static + opened-file tabs with dedupe/cap-8/close), per-session tab & artifact state (switchSession wiring), tool cards retained under Artifacts until P3.2 re-scopes registration. Verified: GUI 193/193 ✓ (incl. 5 new smoke tests), pytest 694 ✓, doc guard 3/3 ✓, import + --help ✓. CI test run green.

[pm25coder]

I tested PR #664 end-to-end on branch head 9cac295 (Windows) — the workspace panel P2 framework works well.

Verification results

• uv run pytest tests/ → 639 passed, 55 skipped (694 total; unchanged from master, no Python changes — as expected)
• cd emrg/gui && npm test → 193 tests, 192 pass, 0 fail, 1 cancelled. The 1 cancelled is the pre-existing P2 onRecovered timing flake in conn-manager.test.js:488 ("Promise resolution is still pending..."), which also cancels on master and passes in isolation — unrelated to this PR. All 5 new P2 framework tests pass.
• node --check on all touched JS files: clean. Import + CLI checks: OK.
• CI: gh pr checks 664 → test PASS (1m8s); PR MERGEABLE.
• Doc counts in Agent.md / README.md / README.cn.md (193 = 43 daemon_client + 19 conn-manager + 22 app-commands + 72 renderer smoke + 15 i18n + 7 integration + 3 commands + 5 build-config + 7 gui-state) match the actual test output; the doc-count guard is green.

Code review notes (non-blocking)

1. Background-session artifact cards render into the active session's pane (cross-session DOM leak).addToolResult(data, sid) correctly stores into the per-sid artifactsBySid bucket, but renderCard(data) unconditionally renders into the shared #result-list DOM. So a tool_finished from a background session paints its card into the currently-active session's artifacts pane, and switchSession() does not re-render the pane from the per-session bucket. openFileTab/closeFileTab already gate DOM writes on (sid || null) === currentSid — applying the same gate to the card render (always store to the bucket, render only when the event sid is current) would make the per-session isolation claim consistent for artifacts too. P3.2 can then consume the per-sid arrays for real file registration.

2. Two different caps for the same artifact list. DOM cards are still capped at MAX_ITEMS = 50 (P1 legacy) while the per-sid artifactsBySid buckets cap at 100. Worth aligning when P3.2 replaces the card rendering.

3. Minor robustness nit (no change needed).#result-resizer's initial CSS right: 277px mirrors DEFAULT_WIDTH(280) − 3; it's always corrected by updateResizerPos() on init, so it only affects the initial paint — fine as-is.

Everything else looks clean: the absolutely-positioned resizer stays out of the #app flex flow (R6-②), .dragging correctly suppresses the width transition (R1-①), collapse/expand correctly keep panelWidth and collapsed persistence separate, and the tab-eviction math in the tests matches the implementation (open 3 → push 10 → first tab /tmp/f3.js).

Nice work — the framework lands cleanly and gives the P3 viewer a solid foundation to plug into.

[argszero]

✅ LGTM — cycle

Reviewed P2 framework (resizable panel + tab bar + per-session tab state) in a local worktree:

• Tab bar (files/artifacts static tabs + open-file tabbar with dedup, 8-tab cap with oldest eviction, close button) + per-session state isolation (openedTabsBySid/artifactsBySid/activeTabBySid, switchSession wired into app.js session switch/new).
• Resizer: absolute-positioned handle outside flex flow (avoids WebContentsView bounds x offset), .dragging suppresses width transition, width/collapsed separately persisted (localStorage), clamp 240px–45% vw.
• Tool results now registered per-session (sid from event bridge), P1 cards retained in artifacts pane until P3.2.
• Verification: renderer smoke 72/72 (incl. 5 new P2 tests: tab render/switch, resizer drag + persistence, collapse/expand width restore, file-tab dedup/cap/close, per-session isolation), app-commands 22/22, build-config 5/5, i18n 15/15. Doc counts consistent (GUI 187→193). (daemon-client/integration env-dependent tests need .venv — not affected by this renderer-only PR.)
  No issues found.

[argszero]

✅ LGTM — cycle (3rd)

Re-verified against branch head 9cac295: resizable panel (absolute resizer, .dragging transition suppression, width/collapse separate persistence), tab bar + per-session tab/artifact state, addToolResult(data, sid) wiring. Local: GUI 193/193 ✓ pytest 694 ✓ doc guard 3/3 ✓. CI test run green (31461002024). Merge conditions met (3 consecutive ✅ from different cycles).

[argszero]

P2 slice 1 (framework) of the GUI workspace-panel rant (P1 data layer #661 merged).

P2.1 Panel structure

• Tab bar: static Files / Artifacts tabs + opened-file tab strip (#result-tabbar, P3 viewer wires in)
• #result-resizer absolutely positioned drag handle (out of the #app flex flow, R6-② — prevents WebContentsView bounds x offset); drag sets panel width, .dragging class suppresses the width transition (R1-①)
• Collapse (⌘) keeps the 40px narrow bar (emrg: keep result-panel toggle visible when collapsed (rant 2026-08-10T14:11:18) #615) with width/collapse persisted separately (two localStorage keys) — expanding restores the dragged width, not the default

P2.2 Tab manager (per-session container mode, gap 5)

• openFileTab/closeFileTab/activateTab — same-path dedupe, max 8 tabs (evicts oldest), closeable tabs, active highlight
• openedTabsBySid / artifactsBySid state maps; switchSession(sid) shows each session's own tabs (wired into app.js switchSession/newSession); addToolResult(data, sid) classifies artifact records per event sid
• Tool-output cards (WorkBuddy P1) intentionally retained under the Artifacts pane — removing them before P3.2's write/edit artifact registration exists would leave the panel empty across merge cycles

i18n: result.tabFiles/tabArtifacts/filesEmpty/resize zh/en.

Tests: +5 renderer smoke (tab render/switch, resizer drag + persistence + .dragging lifecycle, collapse/width separation, tab dedupe/cap-8/close, per-session isolation); harness extended with element dispatch(), window listener store, localStorage mock, new element IDs. GUI 188 → 193; pytest 694 ✓; doc guard 3/3 ✓ (README/README.cn/Agent).

Not in this slice (next): WebContentsView framework (P2.3, parallel with P3.4 HTML preview), file browser tree (P3.1), artifact list re-scope (P3.2), viewer tabs (P3.3).

[argszero]

✅ LGTM — cycle

P2 framework slice of the workspace-panel rant: resizable panel (absolute resizer outside flex flow, .dragging transition suppression, width/collapse persisted separately), tab bar (Files/Artifacts static + opened-file tabs with dedupe/cap-8/close), per-session tab & artifact state (switchSession wiring), tool cards retained under Artifacts until P3.2 re-scopes registration. Verified: GUI 193/193 ✓ (incl. 5 new smoke tests), pytest 694 ✓, doc guard 3/3 ✓, import + --help ✓. CI test run green.

[pm25coder]

I tested PR #664 end-to-end on branch head 9cac295 (Windows) — the workspace panel P2 framework works well.

Verification results

• uv run pytest tests/ → 639 passed, 55 skipped (694 total; unchanged from master, no Python changes — as expected)
• cd emrg/gui && npm test → 193 tests, 192 pass, 0 fail, 1 cancelled. The 1 cancelled is the pre-existing P2 onRecovered timing flake in conn-manager.test.js:488 ("Promise resolution is still pending..."), which also cancels on master and passes in isolation — unrelated to this PR. All 5 new P2 framework tests pass.
• node --check on all touched JS files: clean. Import + CLI checks: OK.
• CI: gh pr checks 664 → test PASS (1m8s); PR MERGEABLE.
• Doc counts in Agent.md / README.md / README.cn.md (193 = 43 daemon_client + 19 conn-manager + 22 app-commands + 72 renderer smoke + 15 i18n + 7 integration + 3 commands + 5 build-config + 7 gui-state) match the actual test output; the doc-count guard is green.

Code review notes (non-blocking)

1. Background-session artifact cards render into the active session's pane (cross-session DOM leak).addToolResult(data, sid) correctly stores into the per-sid artifactsBySid bucket, but renderCard(data) unconditionally renders into the shared #result-list DOM. So a tool_finished from a background session paints its card into the currently-active session's artifacts pane, and switchSession() does not re-render the pane from the per-session bucket. openFileTab/closeFileTab already gate DOM writes on (sid || null) === currentSid — applying the same gate to the card render (always store to the bucket, render only when the event sid is current) would make the per-session isolation claim consistent for artifacts too. P3.2 can then consume the per-sid arrays for real file registration.

2. Two different caps for the same artifact list. DOM cards are still capped at MAX_ITEMS = 50 (P1 legacy) while the per-sid artifactsBySid buckets cap at 100. Worth aligning when P3.2 replaces the card rendering.

3. Minor robustness nit (no change needed).#result-resizer's initial CSS right: 277px mirrors DEFAULT_WIDTH(280) − 3; it's always corrected by updateResizerPos() on init, so it only affects the initial paint — fine as-is.

Everything else looks clean: the absolutely-positioned resizer stays out of the #app flex flow (R6-②), .dragging correctly suppresses the width transition (R1-①), collapse/expand correctly keep panelWidth and collapsed persistence separate, and the tab-eviction math in the tests matches the implementation (open 3 → push 10 → first tab /tmp/f3.js).

Nice work — the framework lands cleanly and gives the P3 viewer a solid foundation to plug into.

[argszero]

✅ LGTM — cycle

Reviewed P2 framework (resizable panel + tab bar + per-session tab state) in a local worktree:

• Tab bar (files/artifacts static tabs + open-file tabbar with dedup, 8-tab cap with oldest eviction, close button) + per-session state isolation (openedTabsBySid/artifactsBySid/activeTabBySid, switchSession wired into app.js session switch/new).
• Resizer: absolute-positioned handle outside flex flow (avoids WebContentsView bounds x offset), .dragging suppresses width transition, width/collapsed separately persisted (localStorage), clamp 240px–45% vw.
• Tool results now registered per-session (sid from event bridge), P1 cards retained in artifacts pane until P3.2.
• Verification: renderer smoke 72/72 (incl. 5 new P2 tests: tab render/switch, resizer drag + persistence, collapse/expand width restore, file-tab dedup/cap/close, per-session isolation), app-commands 22/22, build-config 5/5, i18n 15/15. Doc counts consistent (GUI 187→193). (daemon-client/integration env-dependent tests need .venv — not affected by this renderer-only PR.)
  No issues found.

[argszero]

✅ LGTM — cycle (3rd)

Re-verified against branch head 9cac295: resizable panel (absolute resizer, .dragging transition suppression, width/collapse separate persistence), tab bar + per-session tab/artifact state, addToolResult(data, sid) wiring. Local: GUI 193/193 ✓ pytest 694 ✓ doc guard 3/3 ✓. CI test run green (31461002024). Merge conditions met (3 consecutive ✅ from different cycles).

[argszero]

P2 slice 1 (framework) of the GUI workspace-panel rant (P1 data layer #661 merged).

P2.1 Panel structure

• Tab bar: static Files / Artifacts tabs + opened-file tab strip (#result-tabbar, P3 viewer wires in)
• #result-resizer absolutely positioned drag handle (out of the #app flex flow, R6-② — prevents WebContentsView bounds x offset); drag sets panel width, .dragging class suppresses the width transition (R1-①)
• Collapse (⌘) keeps the 40px narrow bar (emrg: keep result-panel toggle visible when collapsed (rant 2026-08-10T14:11:18) #615) with width/collapse persisted separately (two localStorage keys) — expanding restores the dragged width, not the default

P2.2 Tab manager (per-session container mode, gap 5)

• openFileTab/closeFileTab/activateTab — same-path dedupe, max 8 tabs (evicts oldest), closeable tabs, active highlight
• openedTabsBySid / artifactsBySid state maps; switchSession(sid) shows each session's own tabs (wired into app.js switchSession/newSession); addToolResult(data, sid) classifies artifact records per event sid
• Tool-output cards (WorkBuddy P1) intentionally retained under the Artifacts pane — removing them before P3.2's write/edit artifact registration exists would leave the panel empty across merge cycles

i18n: result.tabFiles/tabArtifacts/filesEmpty/resize zh/en.

Tests: +5 renderer smoke (tab render/switch, resizer drag + persistence + .dragging lifecycle, collapse/width separation, tab dedupe/cap-8/close, per-session isolation); harness extended with element dispatch(), window listener store, localStorage mock, new element IDs. GUI 188 → 193; pytest 694 ✓; doc guard 3/3 ✓ (README/README.cn/Agent).

Not in this slice (next): WebContentsView framework (P2.3, parallel with P3.4 HTML preview), file browser tree (P3.1), artifact list re-scope (P3.2), viewer tabs (P3.3).

[argszero]

✅ LGTM — cycle

P2 framework slice of the workspace-panel rant: resizable panel (absolute resizer outside flex flow, .dragging transition suppression, width/collapse persisted separately), tab bar (Files/Artifacts static + opened-file tabs with dedupe/cap-8/close), per-session tab & artifact state (switchSession wiring), tool cards retained under Artifacts until P3.2 re-scopes registration. Verified: GUI 193/193 ✓ (incl. 5 new smoke tests), pytest 694 ✓, doc guard 3/3 ✓, import + --help ✓. CI test run green.

[pm25coder]

I tested PR #664 end-to-end on branch head 9cac295 (Windows) — the workspace panel P2 framework works well.

Verification results

• uv run pytest tests/ → 639 passed, 55 skipped (694 total; unchanged from master, no Python changes — as expected)
• cd emrg/gui && npm test → 193 tests, 192 pass, 0 fail, 1 cancelled. The 1 cancelled is the pre-existing P2 onRecovered timing flake in conn-manager.test.js:488 ("Promise resolution is still pending..."), which also cancels on master and passes in isolation — unrelated to this PR. All 5 new P2 framework tests pass.
• node --check on all touched JS files: clean. Import + CLI checks: OK.
• CI: gh pr checks 664 → test PASS (1m8s); PR MERGEABLE.
• Doc counts in Agent.md / README.md / README.cn.md (193 = 43 daemon_client + 19 conn-manager + 22 app-commands + 72 renderer smoke + 15 i18n + 7 integration + 3 commands + 5 build-config + 7 gui-state) match the actual test output; the doc-count guard is green.

Code review notes (non-blocking)

1. Background-session artifact cards render into the active session's pane (cross-session DOM leak).addToolResult(data, sid) correctly stores into the per-sid artifactsBySid bucket, but renderCard(data) unconditionally renders into the shared #result-list DOM. So a tool_finished from a background session paints its card into the currently-active session's artifacts pane, and switchSession() does not re-render the pane from the per-session bucket. openFileTab/closeFileTab already gate DOM writes on (sid || null) === currentSid — applying the same gate to the card render (always store to the bucket, render only when the event sid is current) would make the per-session isolation claim consistent for artifacts too. P3.2 can then consume the per-sid arrays for real file registration.

2. Two different caps for the same artifact list. DOM cards are still capped at MAX_ITEMS = 50 (P1 legacy) while the per-sid artifactsBySid buckets cap at 100. Worth aligning when P3.2 replaces the card rendering.

3. Minor robustness nit (no change needed).#result-resizer's initial CSS right: 277px mirrors DEFAULT_WIDTH(280) − 3; it's always corrected by updateResizerPos() on init, so it only affects the initial paint — fine as-is.

Everything else looks clean: the absolutely-positioned resizer stays out of the #app flex flow (R6-②), .dragging correctly suppresses the width transition (R1-①), collapse/expand correctly keep panelWidth and collapsed persistence separate, and the tab-eviction math in the tests matches the implementation (open 3 → push 10 → first tab /tmp/f3.js).

Nice work — the framework lands cleanly and gives the P3 viewer a solid foundation to plug into.

[argszero]

✅ LGTM — cycle

Reviewed P2 framework (resizable panel + tab bar + per-session tab state) in a local worktree:

• Tab bar (files/artifacts static tabs + open-file tabbar with dedup, 8-tab cap with oldest eviction, close button) + per-session state isolation (openedTabsBySid/artifactsBySid/activeTabBySid, switchSession wired into app.js session switch/new).
• Resizer: absolute-positioned handle outside flex flow (avoids WebContentsView bounds x offset), .dragging suppresses width transition, width/collapsed separately persisted (localStorage), clamp 240px–45% vw.
• Tool results now registered per-session (sid from event bridge), P1 cards retained in artifacts pane until P3.2.
• Verification: renderer smoke 72/72 (incl. 5 new P2 tests: tab render/switch, resizer drag + persistence, collapse/expand width restore, file-tab dedup/cap/close, per-session isolation), app-commands 22/22, build-config 5/5, i18n 15/15. Doc counts consistent (GUI 187→193). (daemon-client/integration env-dependent tests need .venv — not affected by this renderer-only PR.)
  No issues found.

[argszero]

✅ LGTM — cycle (3rd)

Re-verified against branch head 9cac295: resizable panel (absolute resizer, .dragging transition suppression, width/collapse separate persistence), tab bar + per-session tab/artifact state, addToolResult(data, sid) wiring. Local: GUI 193/193 ✓ pytest 694 ✓ doc guard 3/3 ✓. CI test run green (31461002024). Merge conditions met (3 consecutive ✅ from different cycles).

[argszero]

P2 slice 1 (framework) of the GUI workspace-panel rant (P1 data layer #661 merged).

P2.1 Panel structure

• Tab bar: static Files / Artifacts tabs + opened-file tab strip (#result-tabbar, P3 viewer wires in)
• #result-resizer absolutely positioned drag handle (out of the #app flex flow, R6-② — prevents WebContentsView bounds x offset); drag sets panel width, .dragging class suppresses the width transition (R1-①)
• Collapse (⌘) keeps the 40px narrow bar (emrg: keep result-panel toggle visible when collapsed (rant 2026-08-10T14:11:18) #615) with width/collapse persisted separately (two localStorage keys) — expanding restores the dragged width, not the default

P2.2 Tab manager (per-session container mode, gap 5)

• openFileTab/closeFileTab/activateTab — same-path dedupe, max 8 tabs (evicts oldest), closeable tabs, active highlight
• openedTabsBySid / artifactsBySid state maps; switchSession(sid) shows each session's own tabs (wired into app.js switchSession/newSession); addToolResult(data, sid) classifies artifact records per event sid
• Tool-output cards (WorkBuddy P1) intentionally retained under the Artifacts pane — removing them before P3.2's write/edit artifact registration exists would leave the panel empty across merge cycles

i18n: result.tabFiles/tabArtifacts/filesEmpty/resize zh/en.

Tests: +5 renderer smoke (tab render/switch, resizer drag + persistence + .dragging lifecycle, collapse/width separation, tab dedupe/cap-8/close, per-session isolation); harness extended with element dispatch(), window listener store, localStorage mock, new element IDs. GUI 188 → 193; pytest 694 ✓; doc guard 3/3 ✓ (README/README.cn/Agent).

Not in this slice (next): WebContentsView framework (P2.3, parallel with P3.4 HTML preview), file browser tree (P3.1), artifact list re-scope (P3.2), viewer tabs (P3.3).

[argszero]

✅ LGTM — cycle

P2 framework slice of the workspace-panel rant: resizable panel (absolute resizer outside flex flow, .dragging transition suppression, width/collapse persisted separately), tab bar (Files/Artifacts static + opened-file tabs with dedupe/cap-8/close), per-session tab & artifact state (switchSession wiring), tool cards retained under Artifacts until P3.2 re-scopes registration. Verified: GUI 193/193 ✓ (incl. 5 new smoke tests), pytest 694 ✓, doc guard 3/3 ✓, import + --help ✓. CI test run green.

[pm25coder]

I tested PR #664 end-to-end on branch head 9cac295 (Windows) — the workspace panel P2 framework works well.

Verification results

• uv run pytest tests/ → 639 passed, 55 skipped (694 total; unchanged from master, no Python changes — as expected)
• cd emrg/gui && npm test → 193 tests, 192 pass, 0 fail, 1 cancelled. The 1 cancelled is the pre-existing P2 onRecovered timing flake in conn-manager.test.js:488 ("Promise resolution is still pending..."), which also cancels on master and passes in isolation — unrelated to this PR. All 5 new P2 framework tests pass.
• node --check on all touched JS files: clean. Import + CLI checks: OK.
• CI: gh pr checks 664 → test PASS (1m8s); PR MERGEABLE.
• Doc counts in Agent.md / README.md / README.cn.md (193 = 43 daemon_client + 19 conn-manager + 22 app-commands + 72 renderer smoke + 15 i18n + 7 integration + 3 commands + 5 build-config + 7 gui-state) match the actual test output; the doc-count guard is green.

Code review notes (non-blocking)

1. Background-session artifact cards render into the active session's pane (cross-session DOM leak).addToolResult(data, sid) correctly stores into the per-sid artifactsBySid bucket, but renderCard(data) unconditionally renders into the shared #result-list DOM. So a tool_finished from a background session paints its card into the currently-active session's artifacts pane, and switchSession() does not re-render the pane from the per-session bucket. openFileTab/closeFileTab already gate DOM writes on (sid || null) === currentSid — applying the same gate to the card render (always store to the bucket, render only when the event sid is current) would make the per-session isolation claim consistent for artifacts too. P3.2 can then consume the per-sid arrays for real file registration.

2. Two different caps for the same artifact list. DOM cards are still capped at MAX_ITEMS = 50 (P1 legacy) while the per-sid artifactsBySid buckets cap at 100. Worth aligning when P3.2 replaces the card rendering.

3. Minor robustness nit (no change needed).#result-resizer's initial CSS right: 277px mirrors DEFAULT_WIDTH(280) − 3; it's always corrected by updateResizerPos() on init, so it only affects the initial paint — fine as-is.

Everything else looks clean: the absolutely-positioned resizer stays out of the #app flex flow (R6-②), .dragging correctly suppresses the width transition (R1-①), collapse/expand correctly keep panelWidth and collapsed persistence separate, and the tab-eviction math in the tests matches the implementation (open 3 → push 10 → first tab /tmp/f3.js).

Nice work — the framework lands cleanly and gives the P3 viewer a solid foundation to plug into.

[argszero]

✅ LGTM — cycle

Reviewed P2 framework (resizable panel + tab bar + per-session tab state) in a local worktree:

• Tab bar (files/artifacts static tabs + open-file tabbar with dedup, 8-tab cap with oldest eviction, close button) + per-session state isolation (openedTabsBySid/artifactsBySid/activeTabBySid, switchSession wired into app.js session switch/new).
• Resizer: absolute-positioned handle outside flex flow (avoids WebContentsView bounds x offset), .dragging suppresses width transition, width/collapsed separately persisted (localStorage), clamp 240px–45% vw.
• Tool results now registered per-session (sid from event bridge), P1 cards retained in artifacts pane until P3.2.
• Verification: renderer smoke 72/72 (incl. 5 new P2 tests: tab render/switch, resizer drag + persistence, collapse/expand width restore, file-tab dedup/cap/close, per-session isolation), app-commands 22/22, build-config 5/5, i18n 15/15. Doc counts consistent (GUI 187→193). (daemon-client/integration env-dependent tests need .venv — not affected by this renderer-only PR.)
  No issues found.

[argszero]

✅ LGTM — cycle (3rd)

Re-verified against branch head 9cac295: resizable panel (absolute resizer, .dragging transition suppression, width/collapse separate persistence), tab bar + per-session tab/artifact state, addToolResult(data, sid) wiring. Local: GUI 193/193 ✓ pytest 694 ✓ doc guard 3/3 ✓. CI test run green (31461002024). Merge conditions met (3 consecutive ✅ from different cycles).

[argszero]

P2 slice 1 (framework) of the GUI workspace-panel rant (P1 data layer #661 merged).

P2.1 Panel structure

• Tab bar: static Files / Artifacts tabs + opened-file tab strip (#result-tabbar, P3 viewer wires in)
• #result-resizer absolutely positioned drag handle (out of the #app flex flow, R6-② — prevents WebContentsView bounds x offset); drag sets panel width, .dragging class suppresses the width transition (R1-①)
• Collapse (⌘) keeps the 40px narrow bar (emrg: keep result-panel toggle visible when collapsed (rant 2026-08-10T14:11:18) #615) with width/collapse persisted separately (two localStorage keys) — expanding restores the dragged width, not the default

P2.2 Tab manager (per-session container mode, gap 5)

• openFileTab/closeFileTab/activateTab — same-path dedupe, max 8 tabs (evicts oldest), closeable tabs, active highlight
• openedTabsBySid / artifactsBySid state maps; switchSession(sid) shows each session's own tabs (wired into app.js switchSession/newSession); addToolResult(data, sid) classifies artifact records per event sid
• Tool-output cards (WorkBuddy P1) intentionally retained under the Artifacts pane — removing them before P3.2's write/edit artifact registration exists would leave the panel empty across merge cycles

i18n: result.tabFiles/tabArtifacts/filesEmpty/resize zh/en.

Tests: +5 renderer smoke (tab render/switch, resizer drag + persistence + .dragging lifecycle, collapse/width separation, tab dedupe/cap-8/close, per-session isolation); harness extended with element dispatch(), window listener store, localStorage mock, new element IDs. GUI 188 → 193; pytest 694 ✓; doc guard 3/3 ✓ (README/README.cn/Agent).

Not in this slice (next): WebContentsView framework (P2.3, parallel with P3.4 HTML preview), file browser tree (P3.1), artifact list re-scope (P3.2), viewer tabs (P3.3).

[argszero]

✅ LGTM — cycle

P2 framework slice of the workspace-panel rant: resizable panel (absolute resizer outside flex flow, .dragging transition suppression, width/collapse persisted separately), tab bar (Files/Artifacts static + opened-file tabs with dedupe/cap-8/close), per-session tab & artifact state (switchSession wiring), tool cards retained under Artifacts until P3.2 re-scopes registration. Verified: GUI 193/193 ✓ (incl. 5 new smoke tests), pytest 694 ✓, doc guard 3/3 ✓, import + --help ✓. CI test run green.

[pm25coder]

I tested PR #664 end-to-end on branch head 9cac295 (Windows) — the workspace panel P2 framework works well.

Verification results

• uv run pytest tests/ → 639 passed, 55 skipped (694 total; unchanged from master, no Python changes — as expected)
• cd emrg/gui && npm test → 193 tests, 192 pass, 0 fail, 1 cancelled. The 1 cancelled is the pre-existing P2 onRecovered timing flake in conn-manager.test.js:488 ("Promise resolution is still pending..."), which also cancels on master and passes in isolation — unrelated to this PR. All 5 new P2 framework tests pass.
• node --check on all touched JS files: clean. Import + CLI checks: OK.
• CI: gh pr checks 664 → test PASS (1m8s); PR MERGEABLE.
• Doc counts in Agent.md / README.md / README.cn.md (193 = 43 daemon_client + 19 conn-manager + 22 app-commands + 72 renderer smoke + 15 i18n + 7 integration + 3 commands + 5 build-config + 7 gui-state) match the actual test output; the doc-count guard is green.

Code review notes (non-blocking)

1. Background-session artifact cards render into the active session's pane (cross-session DOM leak).addToolResult(data, sid) correctly stores into the per-sid artifactsBySid bucket, but renderCard(data) unconditionally renders into the shared #result-list DOM. So a tool_finished from a background session paints its card into the currently-active session's artifacts pane, and switchSession() does not re-render the pane from the per-session bucket. openFileTab/closeFileTab already gate DOM writes on (sid || null) === currentSid — applying the same gate to the card render (always store to the bucket, render only when the event sid is current) would make the per-session isolation claim consistent for artifacts too. P3.2 can then consume the per-sid arrays for real file registration.

2. Two different caps for the same artifact list. DOM cards are still capped at MAX_ITEMS = 50 (P1 legacy) while the per-sid artifactsBySid buckets cap at 100. Worth aligning when P3.2 replaces the card rendering.

3. Minor robustness nit (no change needed).#result-resizer's initial CSS right: 277px mirrors DEFAULT_WIDTH(280) − 3; it's always corrected by updateResizerPos() on init, so it only affects the initial paint — fine as-is.

Everything else looks clean: the absolutely-positioned resizer stays out of the #app flex flow (R6-②), .dragging correctly suppresses the width transition (R1-①), collapse/expand correctly keep panelWidth and collapsed persistence separate, and the tab-eviction math in the tests matches the implementation (open 3 → push 10 → first tab /tmp/f3.js).

Nice work — the framework lands cleanly and gives the P3 viewer a solid foundation to plug into.

[argszero]

✅ LGTM — cycle

Reviewed P2 framework (resizable panel + tab bar + per-session tab state) in a local worktree:

• Tab bar (files/artifacts static tabs + open-file tabbar with dedup, 8-tab cap with oldest eviction, close button) + per-session state isolation (openedTabsBySid/artifactsBySid/activeTabBySid, switchSession wired into app.js session switch/new).
• Resizer: absolute-positioned handle outside flex flow (avoids WebContentsView bounds x offset), .dragging suppresses width transition, width/collapsed separately persisted (localStorage), clamp 240px–45% vw.
• Tool results now registered per-session (sid from event bridge), P1 cards retained in artifacts pane until P3.2.
• Verification: renderer smoke 72/72 (incl. 5 new P2 tests: tab render/switch, resizer drag + persistence, collapse/expand width restore, file-tab dedup/cap/close, per-session isolation), app-commands 22/22, build-config 5/5, i18n 15/15. Doc counts consistent (GUI 187→193). (daemon-client/integration env-dependent tests need .venv — not affected by this renderer-only PR.)
  No issues found.

[argszero]

✅ LGTM — cycle (3rd)

Re-verified against branch head 9cac295: resizable panel (absolute resizer, .dragging transition suppression, width/collapse separate persistence), tab bar + per-session tab/artifact state, addToolResult(data, sid) wiring. Local: GUI 193/193 ✓ pytest 694 ✓ doc guard 3/3 ✓. CI test run green (31461002024). Merge conditions met (3 consecutive ✅ from different cycles).