Add an end-to-end developer guide for extending the native Hermes Desktop
app introduced in #60638: the HermesPlugin contract, PluginContext, every
contribution area (panes, routes, sidebar nav, status/title bar, palette,
keybinds, themes, composer, mount-scoped Contribute), the host API, the
React Query + nanostores data layer, the UI kit + theme variables, the
scoped ctx.rest/ctx.socket backend (plugin_api.py under /api/plugins/<id>)
and its separate enable gate, Settings/defaultEnabled/storage, bundled
plugins, the security model, pitfalls, and a full reference.
- Register the page in the sidebar under Extending -> Plugins.
- Disambiguate from the unrelated web-dashboard plugin SDK from both
directions, and add a map-table row + a desktop user-guide pointer.
- Fill the gaps in the agent-facing hermes-desktop-plugins skill
(ctx.rest/socket + backend, React Query, defaultEnabled, Contribute)
and point it at the new reference so agents know the SDK when writing
addons.
In the Focus layout `files` shares one tab group with `workspace`, so when the
first reply adopts a cwd the reactive files unhide fronted files and yanked the
active tab off the new session (~1s after the reply). #65375 fixed the sibling
"reactive unhide reopens a collapsed side" bug but frontPaneInGroup still stole
the active slot unconditionally.
Only take the active slot when the group's current active pane isn't itself
showable, so a reactive unhide can't steal focus from a pane the user is
viewing while still fronting a valid tab when nothing is shown.
Carries forward linfeng961's diagnosis + fix from #66870, reworked onto the
frontPaneInGroup path introduced by #65375.
Co-authored-by: linfeng961 <133505766+linfeng961@users.noreply.github.com>
* fix(desktop): don't auto-expand user-collapsed side on reactive unhide
When `setTreePaneHidden(paneId, false)` was called for a workspace-gated
pane like `files` (bound to $hasWorkspace), it auto-called
`revealTreePane(paneId)` — which expanded the parent column even when
the user had explicitly collapsed it via Cmd+J.
That meant every session create / resume — anything that flipped
$currentCwd from empty to non-empty — silently re-opened the right
sidebar and persisted that open state to localStorage, so the original
session also showed the sidebar as open on return.
`setTreePaneHidden` is a state primitive; user-intent semantics (open
the side, front the tab) belong to `revealTreePane`. Drop the auto-call
and replace it with a narrower `frontPaneInGroup` helper that only
makes the pane the active tab in its group — visible the next time the
column is opened, without forcing it open now.
* fix(desktop): preserve explicit review reveal
---------
Co-authored-by: David Metcalfe <80915+DavidMetcalfe@users.noreply.github.com>
A manual composer pick stays sticky across new chats (intended), but if that
model later disappears from the provider/config, every new chat kept trying the
dead model and 404'd. Fall back to the profile default in that one case only.
refreshCurrentModel now consults the model-options cache the composer already
populates (no extra fetch): a manual pick is preserved unless manualPickRemoved()
proves it's gone — provider present, non-empty catalog, model absent. An
unknown/absent provider, an empty (re-auth/unconfigured) list, or a not-yet-loaded
catalog all preserve the pick, so a still-valid selection is never clobbered.
Condense the header rationale to the load-bearing invariants and inline
the single-line math-fence guards. No behavior change; scanner logic and
public surface (createScanState / advanceScan / findStableBoundary)
unchanged.
_is_real_user_message no longer accepts a user-role compaction summary as
the human anchor, so _compress_context restores the original user turn
after the summary; update the lifecycle-status test's expectation.
Follow-up hardening on top of the salvaged #66637 commits:
- _insert_real_user_anchor could place the restored human turn directly
next to user-role scaffolding (index-0 insert before a leading synthetic
user turn, or a scaffolding-only transcript), breaking the strict
alternation contract (#55677). Restoration now merges into trailing
scaffolding (anchor text leads, synthetic flags cleared) and appends
after a user-role compaction summary instead of inserting adjacent.
- _is_real_user_message now also rejects user-role compaction summaries
(the compressor pins the summary to role=user when the tail opens with
an assistant turn), so a summary can no longer satisfy the human-anchor
check and skip restoration.
- _latest_user_task_snapshot reuses the same real-user predicate, so the
deterministic task snapshot can no longer anchor on todo snapshots,
truncation notices, or background-process reports.
- The Historical Task Snapshot rewrite keeps the section terminated with
a blank line; the previous replacement consumed the boundary newlines,
gluing the next '## ' heading mid-line and deleting all later sections
on the next iterative compaction.
- Drop _length_continuation_synthetic (no producer anywhere).
- AUTHOR_MAP entry for enzo-adami.
Dusk1e's fix wired _restore_session_cwd into _handle_resume_command, but that
handler was extracted from cli.py into hermes_cli/cli_commands_mixin.py
(094aa85c3) after the PR's base, so the original cli.py hunk no longer applied
(a naive cherry-pick fuzzily misplaced it inside new_session(), where
session_meta is undefined). Transplanted the call to the end of the handler in
its current home; /sessions <id> delegates here so both command forms are
covered. Dusk1e's regression tests carry over unchanged.
Co-authored-by: Dusk1e <yusufalweshdemir@gmail.com>
Rebased onto current main, where _ensure_session_db grew a per-profile
cache (get_hermes_home()-keyed) for /p/<profile>/ multiplex — after this
PR's base. The PR's async rewrite assumed the old single-self._session_db
model, so on current main `session_db=self._session_db` in _create_agent
would pass None in production (the real DB lives in the per-home cache).
Split the concern: keep a SYNC _ensure_session_db (per-profile, used by
_create_agent + the many sync-patching create_agent tests) and add an
async _ensure_session_db_async that captures the profile home on the loop
thread then offloads only the SQLite open via to_thread (single-flight).
Both share _open_and_cache_session_db. Request handlers use the async
variant; _create_agent reverts to the sync call. Updated the first-request
test's FakeDB to accept db_path to match main's SessionDB(db_path=...).
Co-authored-by: necoweb3 <sswdarius@gmail.com>
Surface the session's first-class Project in both chat surfaces: the
Desktop status bar (project name as the workspace label, full cwd in the
tooltip) and the TUI status label + /status output.
One source of truth. The per-profile projects.db is the authority, read
in tui_gateway via _project_info_for_cwd (backed by
projects_db.project_for_path) and threaded through every session.info
emission path the TUI consumes. The Desktop already caches that truth in
$projectTree, so it DERIVES the label from it (projectNameForCwd) instead
of carrying a second per-session $currentProject atom fed from
session.info.
That drops the parallel state #64721 introduced and the entire
reset/reconciliation surface it required (resume, agentless cwd.set,
gateway-switch, fresh-draft): the label is purely derived, so it stays
correct whenever the cwd or the project tree changes. Only explicit,
named projects resolve on both surfaces, so an auto-discovered repo root
keeps the cwd-leaf label everywhere.
Excludes the unrelated markdown shell-fence change bundled in #64721.
- Make create sequence (check + insert + title) atomic via single
_execute_write call with BEGIN IMMEDIATE, closing the TOCTOU window
where two concurrent same-ID creates could both return 201.
- Offload _ensure_session_db() to asyncio.to_thread with single-flight
lock so first-request SQLite init doesn't block the event loop.
- Add concurrent same-ID create test (one 201, one 409) and
first-request path test covering the initialization.
The PR added api_content to get_messages_as_conversation's inline SELECT
but missed the shared _CONVERSATION_ROW_COLUMNS constant used by
get_resume_conversations — _rows_to_conversation references
row["api_content"] but the column wasn't in the SELECT, causing
IndexError on resume-session tests.
Co-authored-by: Soju06 <qlskssk@gmail.com>
Deduplicates the sidecar handling across 9 sites:
- substitute_api_content(): 2 API-bound pop+substitute sites
(chat_completion_helpers, transport — conversation_loop keeps its
inline pop because the current-turn compose fallback needs the value)
- drop_stale_api_content(): 4 rewrite-drop sites
(context_compressor x2, replay_cleanup, agent_runtime_helpers)
- extract_api_content_sidecar(): 3 gateway forwarding sites
(gateway/session, gateway/slash_commands, cli_commands_mixin)
Also: restore the eager _pending_cli_user_message clear on the early
row-creation try (was lost when the crash-persist moved after prefetch —
a crash in compression between the two tries would leave a stale staged
input), and fix a comment indentation nit in run_agent.py.
Co-authored-by: Soju06 <qlskssk@gmail.com>
The first LLM call of every gateway turn gets ~0% provider prompt-cache
hit rate (in-turn calls: 97-99%) because the bytes sent for a turn's
user message are not the bytes replayed next turn: memory-prefetch and
pre_llm_call context are injected into the API copy only, the #48677
persist override writes cleaned content to the DB row, and
get_messages_as_conversation sanitize/strips user and assistant content
on load. Any of these diverges the request prefix at that message and
re-prefills everything after it — measured 27.9s for the first call vs
2.4-5.8s cached at a median ~156k-token context.
Persist what you send: a nullable messages.api_content column stores the
exact content string sent to the API when it differs from the clean
stored content, and replay substitutes it verbatim (no sanitize, no
strip). The injection composition lives in one helper
(turn_context.compose_user_api_content); the turn prologue stamps its
output onto the live user message, the api_messages build sends the
stamped bytes, and every outgoing copy pops the field so it never
reaches a provider. The crash-resilience user-turn persist moves after
prefetch/pre_llm_call so the user row is written once with its final
sidecar; _ensure_db_session stays before preflight compression (session
rotation needs the parent row under PRAGMA foreign_keys=ON). The
current-turn index trackers are re-anchored after compaction rebuilds
the message list, in-place preflight compaction backfills the stamp onto
the already-inserted row, and gateway replay forwards the sidecar only
when the replay pipeline did not rewrite the content. Rewrite paths that
would leave stale bytes (historical image strip, merge-summary-into-
tail, consecutive-user repair merge, stale-confirmation redaction) drop
the sidecar; the chat-completions transport and the max-iterations
summary path strip it defensively. codex_app_server and MoA turns are
excluded from stamping because their wire bytes differ from the
composition. A missing or dropped sidecar degrades to today's behavior
(one cache-boundary miss), never to wrong content.
The busy/priority/monitor/backup×2/drain paths each had near-identical
try/except blocks for transcribe+echo with only log_context, adapter,
and metadata varying. Extract into _transcribe_and_echo_pending_voice
which handles the cache lookup, echo dedup, and exception logging in
one place.
Uses a _UNSET sentinel to distinguish 'caller did not pass metadata'
(use the rich _thread_metadata_for_source fallback) from 'caller
explicitly passed None' (monitor/backup/drain paths that use the
simpler thread_id-only dict).
~120 lines of copy-paste collapsed into one 30-line helper.
merge_pending_message_event extends the existing event's media_urls in
place when two media-bearing messages arrive in quick succession (photo
bursts, consecutive voice messages). The gateway runner caches STT
transcripts on the event via _gateway_pending_stt_text; if the cached
event gains new media after the cache was populated, the stale transcript
was returned instead of transcribing the merged attachments.
Add _invalidate_pending_stt_cache() and call it from both media-merge
branches in merge_pending_message_event so the next transcription call
re-runs against the full merged media list.
Closes the merge-race edge case identified during review of PR #61519.
The function was introduced in d55304c39 but never had a single caller —
verified via git log -S and search_files across the whole repo. The
drain path at _stream_confirmed_final_delivery inlines its logic directly.
Keeping a dead function around that duplicates live logic is a maintenance
hazard: future changes to the live path won't propagate to the dead one.
Extract the inline pure-tool-call tail check to a named helper using
flatten_message_text (canonical content extraction). Fix a SQLite
durability regression: the incremental tool-call persist
(conversation_loop.py:4990) stamps _DB_PERSISTED_MARKER on the assistant
row, so the next _persist_session flush skips it — the filled content
reaches the in-memory transcript but NOT the durable store, and /resume
reloads content="". Pop the marker so the next flush re-writes the row.
Tests pass, ruff clean.
`finalize_turn` guarantees the invariant "delivered final_response =>
assistant row in transcript" (#43849 / #44100) for recovery paths that
return a response without appending a closing assistant message. It
enforces it by checking only the tail's ROLE:
if _tail_role != "assistant":
messages.append({"role": "assistant", "content": final_response})
A tail that is a *pure tool-call turn* — `assistant(tool_calls=[...])`
with no text of its own — satisfies that check while carrying none of the
delivered answer. The append is skipped and the response is never
persisted, so the durable transcript ends at an assistant row the user
never saw as the reply. On the next turn the model replays the user
backlog and re-answers it: exactly the symptom the block was added to
prevent, just reached through a different tail shape.
Observed with the real finalizer: with messages ending
`user -> assistant(content="", tool_calls=[t1])` and
`final_response="Here is your answer."`, the persisted transcript keeps
`content=""` and the answer is absent.
Fill that row's empty content instead of appending. This keeps the
invariant without disturbing the tool-call structure and without creating
an assistant->assistant pair. A tail tool-call row that already carries
model text is left untouched, so no model output is ever overwritten.
Adds regression tests for both: the empty tool-call tail is filled, and a
tool-call tail with existing text is not clobbered.
session.resume built two projections of the same lineage with two separate
get_messages_as_conversation calls — the model-fed copy (tip rows, alternation-
repaired) and the display copy (full lineage, verbatim). The display fetch
already reads a superset of the model fetch (the tip rows are part of the
lineage), so the second call re-scanned the same messages.
Add SessionDB.get_resume_conversations(): one lineage SELECT, split into the tip
(model) and full (display) projections in memory via the extracted
_rows_to_conversation helper. Byte-identical to the two separate reads
(test_get_resume_conversations_matches_separate_reads covers a lineage with a
dangling tool-call tail so repair diverges the two lengths, a single session,
and replayed-user dedup). Both session.resume build paths (deferred + eager) now
use it; the subagent single-read path is unchanged.
From the Desktop performance audit (P1: "Remove repeated resume transcript
work") — the gateway half. The renderer's concurrent REST prefetch is left as-is
(the audit flags dropping it as measure-first).
The sidebar refresh fired three /api/profiles/sessions calls (recents, cron,
messaging), and each one reopened every selected profile's state.db and re-ran
list_sessions_rich + session_count — ~3N DB opens/counts per refresh, on every
turn/broadcast/reconnect.
Add GET /api/profiles/sessions/sidebar: one pass that opens each profile DB
once and runs the three source-scoped queries together (recents scoped to the
active profile; cron + messaging cross-profile), returning the three windows in
one payload. Same read-only projection, 300s active heuristic, and caller-
supplied source taxonomy (recents_exclude / messaging_exclude / source=cron) as
the per-slice endpoint.
Renderer refreshSessions now makes one listSidebarSessions call and distributes
recents/cron/messaging to their stores (cron *jobs* stay a separate getCronJobs
API). Electron splices remote profiles per slice via fetchProfilesSessionSlice
(reusing the proven per-slice merge) so remote correctness is preserved; the
no-remote common case gets the single-open fast path.
From the Desktop performance audit (P1: "Batch sidebar session slices").
The Telegram gateway could go silently deaf for hours: the reconnect ladder
stalled mid-way (e.g. "attempt 4/10, reconnecting in 40s" then nothing) while
the process stayed active(running), so Restart=always never fired.
Root class: every recovery path — the ladder's re-entry
(_schedule_polling_recovery), the pending-update probe (_probe_pending_updates),
and PTB's error callback — gates new recovery on _polling_error_task.done(). If
that single task wedges on any hung await, all recovery returns early forever
and nothing retries.
The heartbeat loop is a separate task, so make it an independent, cause-agnostic
watchdog: if the same recovery task stays in-flight past
_POLLING_ERROR_TASK_STUCK_TIMEOUT (300s — well beyond a healthy ladder attempt's
bounded stop+drain+start+backoff), force a retryable-fatal so the background
reconnector rebuilds the adapter instead of relying on the frozen ladder. This
guarantees progress regardless of *where* the stall is (issue direction #1),
tracked locally so no task-assignment site needs to change.
Also salvages @koduri-mahesh-bhushan-chowdary's #66492 (drain-await timeout),
which closes the one concrete wedge vector documented in the incident
(_drain_polling_connections' unbounded shutdown()/initialize() on a wedged
CLOSE-WAIT pool). The watchdog covers the rest of the class.
Co-authored-by: Koduri Mahesh Bhushan Chowdary <mkoduri73@gmail.com>
_drain_polling_connections() awaited polling_req.shutdown() and
.initialize() without a timeout. When the getUpdates httpx connection is
wedged on a stale CLOSE-WAIT socket, that close can block forever, hanging
_handle_polling_network_error (the tracked _polling_error_task). The task
never completes, so every escalation path — _schedule_polling_recovery,
_probe_pending_updates, the heartbeat verifier — stays gated behind its
in-flight guard, the ladder freezes mid-way, _set_fatal_error is never
reached, and Restart=always never fires: the gateway is alive but silently
dead.
Wrap both drain awaits in asyncio.wait_for with a new module-level
_DRAIN_TIMEOUT (15.0s, matching _UPDATER_STOP_TIMEOUT), mirroring the
existing bounded stop()/start_polling() sites. On timeout the drain logs and
continues, so the handler task completes and the ladder always advances
toward the fatal-restart escalation.
Adds test_reconnect_continues_if_drain_hangs, which wedges the drain and
asserts the handler still reaches start_polling within a hard bound.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
