Add a complete Turkish locale to the WebUI and login page so users can
select Türkçe in Settings, with speech recognition via tr-TR.
Co-authored-by: Cursor <cursoragent@cursor.com>
Add `PATCH /api/mcp/servers/{name}` endpoint that accepts `{"enabled": bool}`,
updates `mcp_servers.<name>.enabled` in config.yaml, and calls `reload_config()`.
Mirrors the existing DELETE pattern.
Also wire the previously-defined-but-unrouted `_handle_mcp_server_delete` into
`handle_delete`, and `_handle_mcp_server_update` into a new `handle_put` +
`do_PUT` in server.py — fixing a pre-existing bug where those handlers existed
but were never reachable over HTTP.
UI: add a toggle button in each MCP server row in the system settings panel
(panels.js). Clicking it calls PATCH and reloads the list. Toggle button is
styled with `.mcp-toggle-enabled` / `.mcp-toggle-disabled` CSS classes. The
`toggle_supported` flag in the list response is now `True`.
i18n: add 5 new keys (`mcp_enable_server`, `mcp_disable_server`,
`mcp_enabled_toast`, `mcp_disabled_toast`, `mcp_toggle_failed`) to all 9
non-English locales (English values as placeholder translations).
Tests: add `TestMcpToggle` class with 7 tests covering disable, enable,
404-not-found, empty name, missing field, response payload, and URL-encoded name.
Update `test_empty_config` and visibility panel assertions to reflect
`toggle_supported: True` and the new toggle button in panels.js.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Fixes#2846. After PR #2758 (the #2653 fix) the update check correctly
falls through to the branch comparison when HEAD has moved past the
latest `v*` tag — so the banner reports the real commit count against
`origin/<branch>`. But `_select_apply_compare_ref` was never updated to
mirror that decision: as long as any `v*` tag exists, it returns
`tags[0]`, even when HEAD is far past it.
Result for everyone running hermes-agent past `v2026.5.16` (i.e. anyone
on agent master between tagged releases):
1. Banner: `Agent (origin/main): 254 updates available` ← correct
2. User clicks Update Now
3. `_select_apply_compare_ref` picks `v2026.5.16` because tags exist
4. `git pull --ff-only origin v2026.5.16` — no-op (HEAD is already past it)
5. `_schedule_restart()` fires anyway, server bounces
6. Next check still reports 254 behind — banner reappears unchanged
`apply_force_update` had the same bug, except worse: `git reset --hard
v2026.5.16` would have actively rewound the user's checkout 254 commits.
The root cause is the same bug class as #2653 — two parallel paths
(`_check_repo_release` and `_select_apply_compare_ref`) that should make
the same decision but didn't. Pre-fix, the "is HEAD past the latest
tag?" predicate lived inline inside `_check_repo_release` only.
Fix
---
Extract `_head_is_past_latest_tag(path, current_tag)` and have both
paths consult it. When HEAD is past the latest tag:
- check path: release check returns None → branch check runs (#2653,
unchanged behaviour, just refactored)
- apply path: falls through to upstream / `origin/<branch>`, never the
stale tag (#2846, new behaviour)
Tests
-----
- `test_select_apply_compare_ref_uses_tag_when_head_is_on_tag` —
unchanged behaviour pinned: HEAD exactly on tag → advance to tag.
- `test_select_apply_compare_ref_falls_through_when_head_is_past_tag` —
the #2846 repro: HEAD = v2026.5.16 + 608 commits → advance to
`origin/main`, not the tag.
- `test_select_apply_compare_ref_no_tags_uses_upstream` — unchanged.
- `test_select_apply_compare_ref_no_tags_no_upstream_uses_default_branch`
— unchanged.
- `test_check_and_apply_paths_agree_when_head_is_past_tag` — symmetry
test, ensures the two paths can't drift apart again.
All 21 tests in `tests/test_updates.py` pass locally (16 existing + 5
new).
Refs #2846, #2653.
Fixes#2853. The `_terminal_shell_preexec_fn` added in `71d8a8fb` called
`prctl(PR_SET_PDEATHSIG, SIGTERM)` so orphaned PTY shells would die when
the WebUI process crashed. But that signal is **per-thread**, not
per-process, and WebUI runs `ThreadingHTTPServer`: every HTTP request is
handled in its own short-lived worker thread.
Flow that broke every Linux user:
1. User clicks the terminal toggle → frontend hits `POST /api/terminal/start`.
2. ThreadingHTTPServer spins up a worker thread to handle that one request.
3. The worker thread calls `subprocess.Popen(..., preexec_fn=...)`.
4. The shell calls `prctl(PR_SET_PDEATHSIG, SIGTERM)` in its preexec_fn.
Its registered "parent" is now the WebUI worker thread that called Popen.
5. The handler returns its JSON response and the worker thread exits.
6. The kernel sees the pdeathsig-parent thread has died and sends SIGTERM
to the PTY shell. The shell dies within ~10 ms of being created.
7. The reader loop sees EIO on the master FD, emits `terminal_closed`, and
the frontend writes `[terminal closed]`.
macOS users were unaffected because `libc.prctl` doesn't exist there —
`ctypes.CDLL(None)` returns a libc handle, `libc.prctl` raises
`AttributeError`, the bare-`except` swallows it, and the shell starts
with no pdeathsig configured.
Empirical verification on this Linux host (real PTY + `subprocess.Popen`
inside a `threading.Thread` that joins immediately):
with preexec_fn → proc.poll() == -15 (SIGTERM), master FD returns EIO
without preexec_fn → proc.poll() == None (alive), master FD returns "HELLO\\r\\n"
Same shell, same PTY, same threading topology as WebUI.
Fix
---
Drop the `preexec_fn` entirely. The orphan-shell-on-crash case the original
PR was navigating is rare for self-hosted single-user installs, and the
existing `atexit.register(close_all_terminals)` + explicit `close_terminal`
paths cover graceful shutdown. A future fix (option B in the issue) can
re-introduce pdeathsig pinned to a long-lived supervisor thread, but that
is a follow-up — this PR is the smallest unbricks-Linux-today change.
Tests
-----
- Invert `test_terminal_shell_uses_parent_death_signal_preexec` →
`test_terminal_shell_does_not_use_pdeathsig_preexec`: asserts
`preexec_fn` is NOT in the Popen kwargs.
- Add `test_pty_shell_survives_when_spawning_thread_exits`: spawns a
real PTY shell via `start_terminal` from a worker thread, waits for
the worker to join, asserts the shell is still alive after a half-second
grace window. This is the contract the original tests never exercised.
- Update `test_terminal_module_registers_graceful_shutdown_reaper` to
refuse re-introduction of the preexec_fn or the `libc.prctl(1, SIGTERM)`
call (treats either as a regression).
All 27 terminal-related tests pass locally.
Refs #2853
Squashed from 2 author commits:
- d2237e23 feat: surface live activity timeline
- eee57ec0 fix: satisfy activity timeline CI guards
Frontend-only telemetry from existing stream events. Replaces empty
Thinking… placeholder with observable run status (Waiting on model /
Waiting on tool result / Working for …). New CSS, new test file.
Right-click any workspace file, folder, or root now shows
'Open in VS Code' alongside the existing Reveal in File Manager action.
- POST /api/file/open-vscode: resolves path via safe_resolve, finds VS
Code via shutil.which() with fallbacks for macOS (/usr/local/bin/code,
app bundle CLI), Linux (/usr/bin/code, /snap/bin/code), and Windows
(%LOCALAPPDATA% and %PROGRAMFILES% user/system installs). Returns a
descriptive error if not found rather than a bare OS error.
- Optional vscode block in config.yaml: command (default: code),
host_path_prefix + container_path_prefix for Docker path mapping.
- i18n: open_in_vscode and open_in_vscode_failed translated in all 10
locales (it, ja, ru, es, de, zh-CN, zh-TW, pt, ko).
- 26 tests in tests/test_2735_open_in_vscode.py covering source wiring,
command resolution, i18n completeness, and live endpoint error paths.
Two confirmed bugs in the thinking/reasoning display:
1. reasoningText was initialized once when the SSE stream opened and never
reset between turns. On the done event, the last assistant message
received the union of every turn's reasoning. Now reset at both turn
boundaries: tool (alongside existing liveReasoningText reset) and
interim_assistant (the other turn boundary where prior reasoning closes).
2. ui.js renderMessages preferred m.reasoning (which could be corrupted by
bug 1) over m.reasoning_content (the clean per-turn value from the
backend). The fallback now reads m.reasoning_content || m.reasoning.
Both fixes are needed: bug 2 alone cannot cover providers that stream
reasoning events without populating reasoning_content on the final API
message.
Updated test_streaming_race_fix.py to scope its reconnect-accumulator
guard to the _wireSSE preamble only, since turn-boundary resets inside
event listeners are intentional and correct.
9 new regression tests in test_issue2565_reasoning_accumulation.py.
Fixes#2713 — live assistant text can truncate at tool-call segment
boundaries during streaming.
Before _resetAssistantSegment() in the tool and interim_assistant SSE
handlers, synchronously flush any pending rAF render work so tokens that
arrived during the 66ms throttle window are written to the DOM before
assistantBody is cleared. Without this flush, the pending _doRender
callback fires after assistantBody is null and skips the write silently,
causing the tail of the pre-tool segment to disappear from the live view.
Implementation:
- Extract _flushPendingSegmentRender() helper (guarded by assistantBody
&& _renderPending) that cancels the pending rAF and synchronously
writes via smd/renderMd/esc — same cascade as _doRender.
- Call the helper from both the tool and interim_assistant handlers
before their respective _resetAssistantSegment() calls.
- Normal cases where the rAF has already fired are unaffected (guard
skips immediately).
Completed transcripts were never affected (renderMessages rebuilds from
the full assistantText accumulator on done).
Adds tests/test_issue2713_streaming_segment_flush.py with 11 static
analysis regression tests pinning the helper shape and call-site
ordering.
Cherry-pick of PR #2796 by @ai-ag2026, squashed from 5 author commits onto current master:
- dcee0563 fix: drop stale optimistic sidebar rows
- 3a73400d fix: clear stale busy state before send
- 46c3b902 fix: preserve server idle rows during optimistic merge
- de51d271 fix: let chat start survive pre-start UI errors
- d2f5c906 fix: hide nonfatal pre-start send warnings
Authorship preserved via --author. Code-only squash (no CHANGELOG).
When current_tag == latest_tag, _check_repo_release returned behind=0
and reported 'Up to date' even if master had moved hundreds of commits
past the tag. This was visible as Agent: v2026.5.16-593-gedb2d9105
alongside a green 'Up to date' pill in Settings.
Run 'git describe --tags --always' after computing behind==0. If the
output includes a -N-gSHA suffix the tag is not at HEAD; return None so
_check_repo_branch runs and counts the real commit gap via rev-list.
When HEAD is exactly on the latest tag the new branch is never taken and
behaviour is unchanged.
Fixes#2653.
Bedrock was silently dropped from the picker because:
1. 'bedrock' absent from _PROVIDER_DISPLAY — group header fell back to
title-cased id; more critically the group fell to the else branch
2. 'bedrock' absent from _PROVIDER_MODELS — else branch has no
auto-detected models, so the group was never appended
3. Fallback env-var detection (hermes_cli unavailable) never checked
AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY
Fix:
- Add 'bedrock': 'AWS Bedrock' to _PROVIDER_DISPLAY
- Add static fallback model list to _PROVIDER_MODELS['bedrock'] with
global Anthropic Claude 4.x cross-region inference profile IDs;
live discovery via hermes_cli.models.provider_model_ids('bedrock')
is used first (existing _read_live_provider_model_ids machinery)
- Detect bedrock in env fallback path when both AWS_ACCESS_KEY_ID and
AWS_SECRET_ACCESS_KEY are present
Tests: tests/test_issue2720_bedrock_model_picker.py (5 new tests)
Closes#2771.
v0.51.117 (PR #2766) introduced a top-level function _inflightStateLimits()
in static/ui.js that collided with the window._inflightStateLimits config
object set in static/boot.js. Because top-level function declarations in
classic (non-module) scripts attach to window, boot.js's assignment
overwrote the function reference, and every later _inflightStateLimits()
call threw TypeError. _compactInflightState() runs on every send(), so
no new chat session could be created — v0.51.117 is effectively unusable.
Reported by @jahilldev, with multiple users (@isma3iloiso, @theDanielJLewis,
@JHVenn) confirming the bug or reverting to v0.51.116.
Fix: rename the function to _getInflightStateLimits() — the window-attached
config key stays under its original name (unchanged for any downstream
code that reads it). Updates all 4 call sites in static/ui.js.
Tests:
- Update tests/test_inflight_storage_quota.py — the existing test
asserted 'function _inflightStateLimits()' in UI_JS as a positive
presence check, which certified the bug. Now asserts the renamed
function name is present AND the old colliding name is absent AND
no stale call sites remain.
- Add tests/test_window_function_collision.py — generalized regression
that scans every static JS file for top-level function declarations
whose name also appears as the target of 'window.X = {...}' or
'window.X = <number>'. This is the exact shape that broke #2715
(_pinnedSessionsLimit in v0.51.106) and #2771. Test fails with a
precise diagnostic naming the file and symbol if the bug class
returns. Confirmed test FAILS on current master (unfixed) and PASSES
on this branch.
Verified end-to-end against the live browser before commit:
- typeof window._inflightStateLimits === 'object' (config preserved)
- typeof window._getInflightStateLimits === 'function'
- _getInflightStateLimits() returns the limits object
- saveInflightState() persists to localStorage without throwing
Full pytest suite: 6308 passed, 6 skipped, 3 xpassed, 8 subtests passed.
Opus advisor: SHIP.
Original PR: #2676 by @lucasrc
Adds POST /api/skills/toggle endpoint that flips skills.disabled in
config.yaml, and a UI toggle in the Skills panel that shows all skills
(including disabled ones) with a per-skill on/off control.
- Backend: new endpoint validates skill exists in filesystem before
toggling. Read-modify-write wrapped in _cfg_lock for thread safety.
Writes through to platform_disabled.webui when present.
- Frontend: each skill-item now has a toggle switch; disabled skills
appear muted but still listed (previously they were filtered out).
- i18n: new toggle keys translated across all 9 non-English locales.
- Tests: round-trip test for disabled list normalization + toggle
endpoint behavior.
Squash-merged from contributor's branch (19 commits + 1 merge commit)
onto current master via the cherry-pick-stale-contributor-prs procedure.
Without --force, git fetch origin --tags refuses to overwrite divergent
local tags and returns 'would clobber existing tag', jamming the entire
WebUI update path indefinitely. The WebUI is a release-tracking consumer
that never pushes tags, so it should always defer to whatever the remote
says a release tag points to. Add --force to all three fetch-tag call
sites:
- _check_repo (the 'Check now' button + periodic check)
- apply_force_update (force-reset to remote HEAD)
- apply_update (stash + pull --ff-only)
Tests:
- Updated 3 existing tests in test_updates.py whose fake_git mocks
asserted the exact ['fetch', 'origin', '--tags'] args list.
- Updated 1 existing test in test_update_banner_fixes.py that asserted
the same shape for apply_update.
- Added 4 new regression tests:
- test_check_repo_fetches_tags_with_force
- test_apply_force_update_fetches_tags_with_force
- test_apply_update_fetches_tags_with_force
- test_check_repo_recovers_from_remote_retag (end-to-end,
proves the bare --tags fetch shape is no longer used)
Closes#2756.
When display.personality is set in config.yaml (e.g. personality: taleb),
new sessions now inherit it automatically instead of starting with
personality=None and requiring an explicit /personality command.
This makes the selected personality sticky across new conversations rather
than requiring per-session activation.
Behavior:
- display.personality values 'none', 'default', 'neutral', '' are treated
as no personality (personality=None), matching TUI gateway semantics.
- Config read is wrapped in try/except — if it fails, personality falls
back to None (no crash, no regression).
- Case-insensitive: 'Taleb' normalizes to 'taleb'.
The /personality slash command still works for per-session overrides as
before; this change only affects the initial default.
Two functions on the /api/session/handoff-summary hot path were opening
sqlite3.connect(...) inside a bare `with` statement, which commits the
transaction at scope exit but does NOT close the connection. Per-turn
invocations accumulated state.db / state.db-wal file descriptors and
CPython heap pages on long-lived worker threads, surfacing as the
multi-GB VmRSS and 6x duplicated state.db fds observed on the live
instance (D0 pre-restart baseline: VmRSS 1,334,248 kB, 55 fds; cold
baseline after restart: VmRSS 136,668 kB, 10 fds).
Wrap both call sites with contextlib.closing(...) (already imported and
used at seven other sites in the same files) so the connection is
closed deterministically:
- api/models.py :: count_conversation_rounds
- api/routes.py :: _persist_handoff_summary_to_state_db
Regression test:
tests/test_issue2233_sqlite_connection_leak.py loops both functions
20 times against a tmp state.db and asserts /proc/<pid>/fd count
does not grow more than 2. Linux-only via sys.platform skip.
D1 live soak against a freshly-built worktree server (port 8799,
isolated HERMES_HOME / HERMES_WEBUI_STATE_DIR) hitting
/api/session/handoff-summary 20 times:
fd_before = 5
fd_after = 5 (growth 0, threshold < 5)
vmrss_before = 52636 kB
vmrss_after = 52636 kB (growth 0 kB, threshold < 30 MB)
The patched fix curve trends below the leak curve.
Rollback: single git revert <this-sha> reverts both file edits.
Refs #2233.
Uğur Murat Altıntas