Add previous messaging session controls

[junjunjunbong]

Summary

• add a setting to show previous messaging sessions that were replaced by reset/compression
• wire the preference through boot/settings/session filtering
• add a separate Hide from list action for imported messaging/CLI sessions without deleting source history

Tests

• node --check static/sessions.js
• node --check static/i18n.js
• python -m pytest tests/test_gateway_sync.py tests/test_1003_preferences_autosave.py

[nesquena-hermes]

@nesquena Ready for independent review when you have a moment. CI is green; staging this for the next batch release once approved.

[nesquena]

Review — end-to-end ✅ (approved after rebase + 3-fix push)

First — thanks for the contribution, @junjunjunbong! The "show previous messaging sessions" toggle is a real UX gap; users with active Discord/Telegram channels currently can't see their reset/compression history without diving into state.db.

What this ships

• New show_previous_messaging_sessions setting (default off) that, when on, bypasses the dedupe/hide logic in _keep_latest_messaging_session_per_source and returns all messaging rows sorted by timestamp.
• Settings panel checkbox + autosave wiring (matches the existing 4-file pattern from feedback_new_settings_pattern).
• A new "Hide from list" menu action on external (messaging/CLI) sessions that calls the existing /api/session/archive endpoint with a clearer toast for imported content.
• 5 new i18n keys.

Note: the branch was stale (rebased onto current master)

PR was opened 2026-05-15 against master @ stage-353 (v0.51.60-era). Master is now at v0.51.89 (~29 releases). I rebased (e8dbf94a) and resolved one keep-both conflict in tests/test_gateway_sync.py where a later email-source test landed in the same insertion gap. CI required workflow approval for the fork — approved, re-ran green on 3.11/3.12/3.13.

Traced against upstream hermes-agent

Pulled a fresh nousresearch/hermes-agent tarball. Grep for _keep_latest_messaging_session_per_source, show_previous_messaging_sessions, and /api/session/archive in hermes_cli/: no CLI counterparts. The setting is purely WebUI sidebar projection; the archive endpoint is webui-only. No cross-tool contract to break. ✓

End-to-end trace

Backend bypass — api/routes.py:2161-2168:

def _keep_latest_messaging_session_per_source(
    sessions: list[dict],
    *,
    show_previous_messaging_sessions: bool = False,
) -> list[dict]:
    if show_previous_messaging_sessions:
        return sorted(sessions, key=_session_sort_timestamp, reverse=True)
    # … existing dedupe + hide-stale logic …

Wiring — api/routes.py:3950-3955 reads settings.get("show_previous_messaging_sessions") and threads it through as a keyword arg.

Setting registration — api/config.py:3928, 4059 — default-False entry in the settings shape and the persisted-fields allow-list (the standard 2-line wiring).

Frontend — static/boot.js exports window._showPreviousMessagingSessions; static/panels.js wires the checkbox into autosave, save-settings, and _applySavedSettingsUi; static/index.html adds the new checkbox row inside Settings → External sessions, just below the existing "Show non-WebUI sessions" toggle.

Hide-from-list action — static/sessions.js:1652-1668 — gated on isExternalSession && !session.archived. Calls POST /api/session/archive with archived: true (same endpoint as the existing "Archive conversation" action), then renders a clearer "Session hidden from list" toast.

Behavioral harness (semantic correctness)

I built a Python harness exercising the helper with the new toggle on/off:

ok    toggle OFF: hide reset/compressed segments (default behavior)
ok    toggle ON: show all segments in timestamp order
ok    toggle ON + invisible active session: still shows all input rows
ok    toggle ON + mixed sources: timestamp-sorted including stale segments
ok    toggle ON + empty rows

All 5 PR-scoped scenarios pass. The bypass cleanly skips both the per-source dedupe AND the gateway-stale-hiding, which is the documented intent.

What I caught — 8 CI test failures

After the rebase, 8 tests failed:

1. tests/test_japanese_locale.py::test_japanese_locale_covers_english_keys — ja missing all 5 new keys
2. tests/test_russian_locale.py::test_russian_locale_covers_english_keys — ru missing all 5 new keys
3. tests/test_spanish_locale.py::test_spanish_locale_covers_english_keys — es missing all 5 new keys
4. tests/test_chinese_locale.py::test_chinese_locale_covers_english_keys — zh missing all 5 new keys
5. tests/test_provider_quota_status.py::test_settings_label_and_description_i18n_keys_exist_for_all_locales — the two settings keys must exist in all 11 locales (locale_count check), but only en+ko had them.
6. tests/test_1466_sidebar_cancel_clarify.py::test_running_sidebar_sessions_get_stop_action — 5200-char window into _openSessionActionMenu no longer covered cancelSessionStream(session) after the 17-line "Hide from list" branch was inserted.
7. tests/test_1466_sidebar_cancel_clarify.py::test_cli_sessions_hide_duplicate_and_delete_in_action_menu — same window issue.
8. tests/test_issue1611_session_profile_filtering.py::test_keep_latest_messaging_runs_after_profile_filter — grep'd for the literal _keep_latest_messaging_session_per_source(scoped) but the call site is now multi-line keyword-arg form.

What I pushed — e8dbf94a

1. i18n parity: Added native-script translations for all 5 new keys to the 9 missing locales (it, ja, ru, es, de, zh-CN, zh-TW, pt, fr). For locales that already use English fallback for related session keys (ru/es/de's session_archive), I still provided localized translations for the new keys to match the project's general translation direction.
2. test_1466 window bump: 5200 → 6400 (precedent: 3200→4400 in #1764, 4400→5200 in #2111). Added the #2294 annotation to the existing window-bump history comment.
3. test_1611 dedupe-position check: loosened the grep from _keep_latest_messaging_session_per_source(scoped) to _keep_latest_messaging_session_per_source( — tolerates both the single-line and keyword-arg call shapes. The position assertion (filter < dedupe) still holds.

Other audit — things that are correct already

• Security: zero new surface. No new endpoints; the "Hide from list" action reuses the existing /api/session/archive POST with the same auth/CSRF gating. Setting is server-validated by the persisted-fields allow-list at api/config.py:4059.
• The 4-file settings pattern is followed exactly: api/config.py (default + allow-list), api/routes.py (read), static/boot.js (export), static/panels.js (autosave wiring). Matches the project convention precisely.
• Autosave parity — tests/test_1003_preferences_autosave.py is updated: 14 → 15 fields, count assertions bumped correctly. The new checkbox settingsShowPreviousMessagingSessions is registered in PREFERENCE_FIELDS_AUTOSAVE and is exercised by _preferencesPayloadFromUi.
• JS syntax: all 4 modified .js files (sessions.js, i18n.js, boot.js, panels.js) parse cleanly in Node after my changes — verified.
• Backwards-compat: default-False preserves prior behaviour exactly. Users who don't flip the toggle see no change. ✓

Edge-case trace

Scenario	toggle	Expected	Actual
Active gateway session visible, stale reset row	OFF	active kept, stale hidden	✅ unchanged from before
Active gateway session visible, stale reset row	ON	both kept, timestamp-sorted	✅
Active session invisible (zero msgs), older history row	ON	older history kept	✅
Multiple sources mixed, some stale	ON	all kept, timestamp-sorted	✅
Empty input	ON	empty output	✅
Non-messaging row (webui)	OFF/ON	always kept (key=None passthrough still applies when toggle off; toggle on returns everything)	✅

Tests

• PR test: test_previous_messaging_setting_keeps_reset_history — passes; asserts both default-off and toggle-on outputs.
• Updated test: test_all_15_preference_fields_have_autosave_payload_entries — passes (14→15 count bumped to match new field).
• All 4 locale-parity tests: pass after fix push.
• Settings i18n parity (all 11 locales): passes after fix push.
• Two _openSessionActionMenu window tests: pass after window bump.
• test_keep_latest_messaging_runs_after_profile_filter: passes after grep loosening.
• Full suite: 5828 passed, 63 skipped, 0 failed in 50.18s.
• CI after fix push: test (3.11/3.12/3.13) all green.

Minor observations (non-blocking)

• "Hide from list" alongside "Archive conversation" is technically redundant — both call POST /api/session/archive with archived: true, both show for external non-archived sessions. The contributor's intent (clearer copy for imported sessions) is reasonable, but external sessions now show TWO menu items that do the same thing with different toast text. Possible cleanup direction: hide "Archive conversation" when "Hide from list" is shown (i.e. external sessions only see "Hide from list"; internal sessions only see "Archive conversation"). Not in scope for this PR.
• Composition with PR #2286: The bypass-on-toggle predates #2286's "scope active gateway sources to visible sessions" fix. When both land, they compose cleanly — bypass runs first when toggle is on; #2286's improved logic runs otherwise. No merge conflict expected since both touch different parts of the function (top vs the middle of the slow path).
• No CHANGELOG entry with the PR. Not blocking — release agent stages typically write these on the merge sweep.

Recommendation

✅ Approved after fix. Parked at approval — ready for the release agent's merge/tag pipeline.

Thanks again for picking this up — the bypass approach is correct (a feature flag that short-circuits the dedupe is exactly the right shape for this user-controlled override), and the 4-file settings wiring + the autosave test bump shows you read the existing patterns carefully. The i18n parity gap was easy to miss for an external contributor; the project's strict parity tests are unusually picky.