Each mid-session `/model` switch appended a `[System: The active model for
this chat has changed to …]` user-role marker to the live conversation history
and never removed the prior one. N switches left N stale markers, all re-sent
to the provider on every subsequent turn — the issue measured ~80-120 tokens
each (Chinese+English), so 5 MoA-preset switches burned ~400-600 context
tokens per API call, permanently. Only the newest marker is meaningful (it
names the currently-active model); the rest are pure waste. Fixes#65891.
`_append_model_switch_marker` now strips any earlier markers from
`session["history"]` (in place, under the existing history_lock) before
appending the new one, so the live payload carries exactly one marker. This is
the payload re-sent each turn, and it's self-healing across resumes: whatever
markers a history reload brings back, the next switch collapses to one.
A stable `_MODEL_SWITCH_MARKER_PREFIX` constant is shared by the builder and
the `_is_model_switch_marker` predicate so the two can't drift; the marker
string itself is byte-for-byte unchanged. Existing behavior (role='user' per
the single db.append_message persistence) is preserved.
Scope note: this dedups the in-memory history (the per-turn cost the issue
quantifies). The DB still persists one row per switch for audit/resume; I did
NOT soft-archive prior marker rows because `active=0, compacted=0` is the
repo's rewind/undo sentinel (hermes_state.py get_messages docs), so
deactivating a marker naively would make it look rewound — a follow-up that
needs its own archive semantics.
Tests (tests/tui_gateway/test_model_switch_marker_role.py, +5): second switch
replaces (not stacks) the marker; the issue's exact 5-MoA-switch sequence
leaves one marker naming the last model; real conversation turns are preserved
in order; a stale marker sitting between turns is stripped; history_version
increments once per switch. `pytest test_model_switch_marker_role.py` → 13
passed (8 existing unchanged). Authored on Windows.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013b1XyXitAxV7phGmKWigJX
Setting prompt_caching.cache_ttl to a falsy value (false, null, off,
disabled, no, none) now fully disables prompt caching instead of
being silently ignored.
The disable propagates through anthropic_prompt_cache_policy() (early
return when _cache_disabled flag is set) and restore_primary_runtime()
(override after snapshot restore), so it survives /model switches and
fallback re-derivation — the gap that caused #56105 to be reverted in
#56126.
Salvage of #33555 by @BB-light, with model-switch/fallback survival
gap fixed on top.
Co-authored-by: BB-light <BB-light@users.noreply.github.com>
The claim 'late tools then require an explicit /reload-mcp' was false on
current main: the between-turns prologue refresh (agent/turn_context.py)
picks up late-connecting servers cache-safely at every turn boundary,
and ACP has no /reload-mcp. The daemon's real marginal value is tool-list
freshness in the [session created -> first message] window.
Review follow-ups on the #32811 salvage:
- Hold state.runtime_lock and bail on is_running so the pre-first-turn
guard can't race the first prompt dispatch (a refresh publishing
mid-turn would swap tools= and break the just-created cache prefix).
Regression test mutation-checked (guard removed -> test fails).
- In-memory-only session lookup in the daemon: get_session() falls
through to a DB restore that builds a whole new AIAgent just to
decide no-op (TUI equivalent also checks its in-memory dict only).
- Pass quiet_mode=True explicitly, matching the TUI/gateway callers.
- Use ensure_mcp_discovery_before_agent_build() (landed on main after
the PR) instead of bare wait_for_mcp_discovery() so the ACP agent
build is self-sufficient and gets the retry-after-zero-connected
allowance, matching CLI/one-shot construction sites.
ACP entry.py fires MCP discovery in a background daemon thread, but
_make_agent snapshots tools once at build and never re-reads the registry.
Unlike CLI/TUI, ACP had no bounded wait before the snapshot and no
late-refresh for configured (config.yaml) MCP servers — a reachable-but-
slow server that finished after agent build was invisible for the whole
session.
Changes:
- acp_adapter/session.py (_make_agent): call wait_for_mcp_discovery()
before AIAgent construction, bounded by mcp_discovery_timeout (default
~1.5s). A dead server can't block; servers that miss the bound are
picked up by the late-refresh below.
- acp_adapter/server.py (_schedule_mcp_late_refresh): new method on
HermesACPAgent — if discovery is still in flight after session creation,
spawns an off-critical-path daemon that joins it (bounded 30s), then
rebuilds the tool snapshot via the shared refresh_agent_mcp_tools helper.
Cache-safe: only runs pre-first-turn (_user_turn_count/_api_call_count
both 0); once the user has sent a message the snapshot is frozen, exactly
as TUI PR #48403 does.
- Called from new_session, load_session, resume_session.
- Mirrors the TUI pattern (tui_gateway _schedule_mcp_late_refresh, PR
#48403) and the CLI pattern (get_tool_definitions → wait_for_mcp_discovery).
Tests:
- Replace the AST-based test (source-text inspection) with three
behavioral regression tests in tests/acp_adapter/test_acp_mcp_discovery.py:
1. Blocked discovery does not block startup (non-blocking contract)
2. Delayed discovery lands tools via late-refresh (pre-first-turn)
3. Late-refresh is cache-safe: skips rebuild after first turn
Addresses teknium1 review on PR #32811.
Fire-and-forget MCP server connections on a daemon thread so the
gateway / CLI / ACP process becomes interactive immediately instead
of blocking on slow remote MCP servers (HTTP timeouts, sluggish
stdio boot). Previously `hermes --tui` waited 2-5 s after the splash
screen before rendering the UI while `discover_mcp_tools()` ran
synchronously on the critical path.
Changes:
- tools/mcp_tool.py: add `discover_mcp_tools_background()` — thin
wrapper that spawns `discover_mcp_tools()` on a named daemon thread
- tui_gateway/entry.py: call `discover_mcp_tools_background()` before
sending gateway.ready (replaces inline call that blocked the JSON-RPC
pipe for the TUI Ink app)
- hermes_cli/main.py:
- skip `\_prepare_agent_startup()` for TUI path — plugins, MCP, and
shell hooks are only needed by the CLI agent loop; the TUI's
gateway subprocess discovers them independently (~370 ms saved)
- fast-path in `\_make_tui_argv()`: when `dist/entry.js` exists and
is fresh, skip npm install / rebuild checks entirely (~350 ms saved)
- cli.py (`\_prepare_deferred_agent_startup`): same background pattern
for deferred startup (Termux interactive CLI)
- acp_adapter/entry.py: same pattern so ACP server launches asyncio
immediately while MCP connects in parallel
Result:
- TUI Python wrapper: ~730 ms → ~80 ms (9× faster)
- gateway.ready: ~2700 ms → ~400 ms (7× faster)
- Total TUI cold start: ~3400 ms → ~480 ms
Related: #29726, #29184, #19326 (closed stale)
Closes#29726
All-fixed splits used to promote the last track to flex-grow and drop its
max clamp. Review and files both declare maxWidth, so ⌘G/⌘J ballooned them
and sash overrides only set a basis that grow still expanded past.
EventBridge initialized each session's last_seen timestamp to 0.0, so
the first poll after 'hermes mcp serve' starts treated every saved
user/assistant message in state.db as a fresh events_poll event.
The fix establishes a per-session timestamp baseline on startup via
_establish_baseline(), recording the latest existing message timestamp
without emitting events. Only messages written after the baseline are
delivered on subsequent polls.
Also hoists _ts_float to module-level (needed by _establish_baseline)
and adds ImportError fallbacks for hermes_constants imports so the
bridge works in environments where the module isn't on the path.
Salvage of #13414 by @afurm, re-applied by @HeLLGURD in #41239.
Co-authored-by: afurm <afurm@users.noreply.github.com>
resolveNewSessionCwd now inherits the focused chat's workspace when you
aren't drilled into a sidebar project, so a new tab or draft stays in
the same repo as the chat you were looking at.
Workspace indicator was stuck on the primary $currentCwd while timers
and context already tracked focus. Resolve from the focused runtime
slice, then the stored session row, with a mid-switch ownership gate.
Path tips across chrome use displayPath (~/).
A $currentCwd listener in the terminal store picks the user tab whose
live shell cwd (restoreCwd, falling back to launch dir) matches the
session's workspace. Selection only: no tab is created, closed, or
revealed; detached sessions and unmatched cwds leave the rail alone,
and an already-matching active tab keeps focus.
Same gesture the pane tabs already carry — isMetaClose hoists from
pane-tab.tsx into lib/middle-click.ts beside its sibling middle-click
gesture, so the two surfaces share one predicate instead of drifting.
A turn that ended without its message.complete (turn crash, reconnect gap,
steer race) left its streaming bubble pending:true forever. The next user
message then landed after it, stranding a live dither indicator
mid-transcript.
Three layers:
- session.info running=false (the agent loop's finally-block signal, the
only settle edge those paths still emit) now finalizes the streaming
bubble via the same math as Stop.
- A fresh submit settles any leftover pending bubble before appending the
new user message, and drops a stale streamId so the new turn seeds fresh.
- AssistantMessage renders the loading/stall indicator only on the thread's
last message, so no upstream state bug can ever paint one mid-transcript.
Guard the early-auth pairing gap: unknown DMs must reach the gateway when
pairing is the effective unauthorized-DM behavior, while ignore mode and
unauthorized group senders stay rejected.
The #40863 intake prefilter rejected unauthorized DMs whenever an allowlist
existed, so gateway pairing never ran even when the operator set
telegram.unauthorized_dm_behavior: pair (which must win over the #9337
allowlist silence default). Pass those DMs through; groups stay blocked.
Non-interactive sessions (hermes chat -q, hermes -z) snapshot the tool
registry at AIAgent construction time. If background MCP discovery hasn't
finished, MCP tools are invisible for the entire session — and unlike
interactive mode, there is no between-turns late-binding refresh to recover.
Root cause: wait_for_mcp_discovery() only joins an already-created discovery
thread, so it no-ops if a direct/single-query path reaches agent construction
before MCP startup created that thread. Oneshot._run_agent() didn't call it
at all.
Fix:
- Add ensure_mcp_discovery_before_agent_build() helper to mcp_startup.py:
idempotently starts discovery if needed + bounded wait. Fail-open on errors.
- Add single_query parameter to _resolve_discovery_timeout/wait_for_mcp_discovery:
uses mcp_single_query_discovery_timeout (default 15s) instead of the
interactive mcp_discovery_timeout (1.5s) because one-shot sessions have no
second turn to recover.
- Wire into CLI _init_agent (single_query from _single_query_mode flag set
in cli.py's single-query path) and oneshot._run_agent (single_query=True).
- Interactive sessions unchanged: keep 1.5s bound (between-turns refresh covers).
Closes#38448, #51316, #37013, #68137
Composite salvage of #60017 (chrishart0), #51322 (Bartok9), #38620 (buptwz),
#43544 (halonke), #36882 (vanhoof).
mirrorSelection called textarea.select() whenever xterm had a scrap,
which replaced any chat highlight so ⌘C copied the wrong thing. Only
claim the document selection while the terminal is focused and nothing
outside it is highlighted.
User bubbles are buttons, so the global user-select:none rule killed
text selection. Right-click-to-react and click-to-edit also ate a live
highlight. Prefer selection when one exists.
get_custom_provider_extra_headers() was returning the result of
normalize_extra_headers() on the first matching base_url, even when
that entry had no extra_headers configured. A later providers.<name>
entry sharing the same URL but with headers set was therefore ignored.
Fix: store the normalized headers and only return when non-empty,
otherwise continue searching the remaining entries.
Fixes#74465
Verify that set_config_value and unset_config_value refuse to write
when config.yaml contains YAML syntax errors, and the original file
is left intact.
set_config_value() and unset_config_value() silently replaced the
entire config with an empty dict when config.yaml could not be
parsed. A single YAML syntax error would cause 'hermes config set'
to wipe all settings and write only the new key. Now exits with
error and preserves the existing file.
The first regeneration kept a ~5% transparent margin around the icon
plate (94.9% coverage). Windows expects the plate itself to be the icon
edge — scale the artwork's rounded plate to span the canvas exactly.
The shipped artwork bakes in the macOS-style ~10% transparent margin
(content covered only ~80% of the canvas), so the taskbar/titlebar icon
rendered visibly smaller than neighboring Windows apps, which draw
full-bleed.
- Regenerate assets/icon.ico full-bleed (~95% coverage) from the same
art, with the standard 16-256px frames. This feeds both the exe stamp
(set-exe-identity via rcedit) and the installer.
- On Windows, resolve the BrowserWindow icon from the full-bleed ico
(resources/icon.ico, shipped via extraResources) before falling back
to the padded apple-touch PNG.
macOS is untouched: the dock icon and icon.icns keep the padded art,
which is correct there.
main reworked the same surface while this was open, so three hunks needed
deciding rather than accepting.
Logs became summon-only (#75862): the contribution only exists while $logsOpen
is on, docked as its OWN zone beside the terminal instead of a tab in its
strip. That supersedes the static logs pane and the bindToolPaneCollapse call
here — main already registers logs' closer/opener directly, so both were
dropped in favour of its version.
main also added a ⌘K "Toggle terminal" row reading $terminalTakeover, and
gave logs back a 7.5rem minHeight under a comment claiming the terminal's
sizing rule. Both are the bugs this branch fixes, so they move onto the
shared behaviour: the palette row reads isPaneVisible/togglePaneVisible like
every other pane toggle, and logs loses the floor so the comment is true —
the sash folds its zone to the rail instead of stranding a sliver.
A turn that rewrote twenty files grew a twenty-row card, so the summary
that is supposed to close the turn became the thing you scroll past to
reach the composer. Cap the rows at ~5 and let the clipped edge fade,
the way every other overflow in the app reads.
The horizontal padding moves onto the scroller so a row's hover fill
still bleeds to the card's edge instead of stopping at a scroll gutter.
The kanban drawer had grown the only edge-aware masked scroller in the
tree, and the next surface that wants one would have copied it. Lift it
to components/ui as FadeScroll, export it on the plugin SDK, and leave
kanban's ScrollFade as a name its call sites already pass `max` to.
The mask math comes out as two pure functions. jsdom's CSS parser drops
any gradient containing calc(), so a rendered mask-image can't be read
back off the style attribute -- edgeMask/scrollEdges are testable for
real where the component's inline style is not.
Google API and authentication packages permit vulnerable httplib2 and pyasn1
transitives, while the Workspace and Google Chat runtime installers previously
treated any importable version as sufficient. Existing environments could
therefore remain vulnerable after the project dependency pins were repaired.
Carry the fixed versions through the Google and Vertex extras, lazy feature
requirements, lockfile, and both runtime installers. Route the documented
Google Chat installation path through its maintained secure requirements
instead of an unconstrained direct pip command.
Detect stale distributions, install only unsatisfied requirements, and verify
the result before continuing. Behavioral tests cover those repair invariants
without freezing manifests, lockfiles, or complete package sets.
Related #72108
Extracted from #72840
Co-authored-by: Teknium <127238744+teknium1@users.noreply.github.com>
The rebase onto main's extracted _sleep_before_write_retry() method left
three call sites pointing at the dropped local helper; rewire them.
Also adds contributors/emails mappings (Dannou, trippyogi, spfcraze).
Installs whose state.db reached schema_version >= 22 before the task
dimension was added carry a 5-column PRIMARY KEY on
session_model_usage. The column reconciler ADDs task as a bare
nullable, but SQLite cannot ALTER a primary key, and the version-gated
v22 rebuild is unreachable (current_version < 22 already false), so
the composite 6-column key never lands. Every upsert in
_record_model_usage then fails with 'ON CONFLICT clause does not match
any PRIMARY KEY or UNIQUE constraint', aborting the enclosing write
transaction — token/cost accounting permanently dead (#73823).
Add an idempotent _heal_session_model_usage_pk() modeled on
_heal_gateway_routing_pk(), run unconditionally from _init_schema on
every open. Salvaged from #73838 with fix-ups:
- ported to SessionSchemaMixin in hermes_state_schema.py (the schema
code moved out of hermes_state.py in 21c7ae8563; the PR targeted the
old location)
- rebuild wrapped in a PRAGMA foreign_keys=OFF/ON window: the
connection enables FKs before _init_schema and OR IGNORE does NOT
suppress FK violations, so a single orphaned usage row (session
pruned while accounting was broken) would have aborted the heal
- COALESCE('') on the nullable reconciler-added task column (and the
billing columns) during the copy
- stale-v22+ regression tests: rebuilt PK + restored upsert, orphan
rows survive the FK window, healthy-DB no-op, no legacy leftover
Fixes#73823
Under dual gateway/agent WAL contention (FTS5 trigram sync holding the
write lock on large appends) the SQLite engine can raise a transient
'no more rows available' error. The exception CLASS varies with the
build — some surface it as InterfaceError, a SIBLING of DatabaseError —
so it escaped both existing retry branches in _execute_write on attempt
0 and killed the turn as session_persistence_failed even though the
identical write succeeds standalone.
Port of #74934 onto the deadline-patience rewrite (8da8a7887d): the
PR's attempt-counted constants (60 retries / 300ms jitter / 2.0s engine
timeout) predate that rewrite and are superseded by the patience
budget, so they are intentionally NOT carried over. Instead the check
is message-scoped and rides the existing deadline/patience loop:
- extract the jittered-sleep-within-deadline logic into a shared
_sleep_before_retry helper (behavior-preserving for locked/busy)
- retry 'no more rows available' from OperationalError, DatabaseError
(checked BEFORE the FTS-corruption rebuild path so it is not
misrouted), and a message-scoped sqlite3.Error catch-all
- any other error in any class propagates untouched on attempt 0
Tests: transient InterfaceError retried to success; unrelated
InterfaceError propagates immediately; DatabaseError variant retried;
exhausted patience surfaces the original error.
The micro-compaction defrag pass (_defrag_rolling_summary) rewrites the
newest MICRO marker's content and pops _DB_PERSISTED_MARKER from the
LIVE dict in place — the same in-place pop class finalize_turn's fill
site was fixed for in #75170. Without invalidation the bounded
flush-scan cursor identity-skips the rewritten marker row and the
defragged rolling summary never reaches state.db (resume rehydrates a
stale summary).
The compressor holds no agent reference, so the pop site raises
_flush_scan_cursor_invalidated and the finalize_turn micro-compaction
block consumes it, setting agent._db_flush_scan_prefix = None.
The module-scope pop sites (context_compressor.py:175/224) operate on
fresh copies — identity-breaking by construction — and need no flag.
Follow-up to #75170 (fix-the-class sweep of _DB_PERSISTED_MARKER
in-place pops).
The bounded flush-scan in _flush_messages_to_session_db_unlocked skips
the identity-matched prefix of its previous snapshot, on the documented
assumption that no code path pops _DB_PERSISTED_MARKER from a live dict
in place. finalize_turn's pure-tool-call-tail fill is exactly that path:
it pops the marker so the filled content gets re-persisted — but the
cursor then skips the row anyway, so the delivered final response never
reaches state.db and /resume replays content="" (the #43849/#44100
class resurfacing via the perf cursor). Invalidate the cursor at the
pop site so the filled row is re-examined.
Rod Boev