* refactor(connectors): cut comments that restate the code
Connector modules (tools/connectors, tui_gateway connector RPCs, desktop
connector card/store) keep only comments that carry a non-derivable why or
a cross-module contract. No behaviour change.
* feat(connectors): managed connect runs on the connection operation
Managed `connect` / `reconnect` mint one ConnectionOperation for every target and, on a
desktop session, block the tool turn until the operation settles; the result is per-target
outcomes and never carries a connect link. Off the desktop the result carries the links and
returns at once (PR3 delivers them as their own message).
Why: the previous leg handed the model a URL and a `wait` verb, and the renderer ran its own
2s poller on top of the backend's 5s one; both walked the whole gateway catalog at two vendor
calls per page to read one row (~3 Composio calls/s per pending target). A hidden composer
message started the model's `wait` on the user's behalf. None of it was observable from the
operation the MCP leg already used.
What the operation looks like now:
- `contract.py`: TargetState / Actor / SettleReason enums and the `(kind, from) -> {to: actor}`
transition table. `operation.transition()` enforces it; a card cannot claim a managed
target `connected`, only the backend watcher can.
- `live.py`: one open operation per session, found by `op_id`. `connectors.operation.status`
reads it, `connection.respond` drives it, `pending_connection` on resume replays it.
- `run.py`: the one lifecycle for both target kinds (prepare -> card -> wake/observe loop ->
settle -> result). The managed `observe` hook polls the gateway list once per tick for the
whole operation; the exact-status route replaces that call when the gateway ships it.
- `connection.update` is emitted on every transition and on settlement; registered in the
shared event contract with the operation vocabulary typed on the TS side.
- `wait`, `_rendered_links`, `_seen_instructions`, the just-minted bounce and `_clamp_timeout`
are deleted. `force` on `reconnect` always reinitiates; plain `reconnect` repairs only what
the gateway reports disconnected.
- `connections.wait_timeout_seconds` is removed from config defaults, the example and the
docs. The deadline is `OPERATION_DEADLINE_SECONDS = 300` in `operation.py`; the key was
added on this unmerged train so no migration is needed.
- Wire model: `statusReason` parsed on connection results; the seven-state `connectionStatus`
is typed on list items and an unknown value fails validation; `CONNECTION_REQUIRED` carries
`connect_card_available` instead of the link when the session platform is `desktop`.
Session platform, not callback presence, decides whether a card exists: the GUI bridge
attaches callbacks to every backend session, terminal TUI included.
* feat(desktop): connector card subscribes to the connection operation
The card renders from the backend's operation instead of driving its own: `connector-flow.ts`
(the renderer's 2s `connectors.list` poller, its 120s client deadline and `keepWaiting`) is
deleted, and both hidden composer submits in `connector-tool.tsx` go with it. The model is
never nudged into a `wait`; the tool call is blocked on the backend until the operation
settles.
- `connection-request.ts` is the operation store: keyed by `op_id`, one entry per session,
`applyOperationStatus` / `applyConnectionUpdate` as pure reducers, `respond` leaves the
entry in place (the backend answers with `connection.update`), `ConnectionTargetOutcome`
is a discriminated union the backend's transition table accepts.
- `input-requests.ts` applies `connection.update`; `connection.expire` and the resume
snapshot correlate by `op_id` (a snapshot has no `request_id`).
- `ConnectorOffer` renders one `ConnectorCard` per target from a single
`Record<ConnectionTargetState, phase>` table; Connect opens the stored link, Try again on
failed / expired reissues through `connectors.connect` on the open operation, Not now is a
per-target `skipped`, Continue settles. A settled operation renders `ConnectorSummary` rows
with no live control.
- `tool-render-class.ts`: `manage_connections` renders the card regardless of
`HERMES_GUEST_ONBOARDING`; the flag still gates the onboarding flow, not the card. The
backend gate already decided admission; a card only exists because the tool was admitted.
- `mcp-setup-tool.tsx` speaks the same outcome vocabulary (connected / skipped / failed).
- `ConnectorRow.connectionStatus` is the seven-state literal union, not `string | null`.
- The guided-onboarding poller (`first-build-connectors.ts`) keeps its own row/phase types
and compiles unchanged; PR3 moves it onto the operation.
anti-slop: no net-new findings (17 touched files vs 11d1a12472).
* fix(connectors): the card never parks the tool thread; every update carries the snapshot
Found by the pre-PR adversarial review and a real-path E2E test (both left in the tree).
- The desktop `connection_callback` was still `_block("connection.request", ...)`, which parked
the tool thread on a private request-id Event until a `_respond` that no longer exists for
this event. `connection.respond` settled the operation but the tool waited its full deadline
before the watcher loop even started. The callback now only emits the card; the operation's
own wake loop is the wait. The MCP leg's blocking bridge goes with it: the card answers
through `connection.respond` like every other card.
- `connection.request` and every `connection.update` frame carry the full target snapshot
(state, link, detail). The initial mint happened before the card existed, so the renderer
never saw the links and Connect stayed disabled; a Continue settlement stamped
`not_connected` on the backend while the card still showed `initiated`. The store now
overlays the snapshot; no state is reconstructed from deltas.
- The `connection.update` emitter is a class-level `on_change` slot on the operation, set
once by `register()` (a second `register()` no longer stacks wrappers); session lookup takes
`_sessions_lock`; a re-minted link on an `initiated` target goes through `refresh_link()`
and emits, instead of a bare attribute write.
- `session.interrupt` is checked before the first observe, so an interrupted call settles
`interrupt`, not `all_resolved`.
- A gateway list reporting `expired` for an initiated target is recorded with actor `clock`
(the contract's owner of that edge); it raised `IllegalTransition` before.
- Dead `keepWaiting` i18n keys from the deleted renderer poller removed.
tests/tui_gateway/test_connector_operation_e2e.py runs the desktop lifecycle through the real
tool, registry, gateway RPC handlers and callback bridge with only the HTTP client faked.
* docs(connectors): prompts and docs describe the operation, not the deleted wait verb
The onboarding prompts told the model to call action="wait" with timeout_seconds and to
expect a hidden [setup]/[connectors] note; both are gone. tool-search.md and
toolsets-reference.md said the model gets a connect link on the desktop. tui_gateway/AGENTS.md
gains the connection-operation row of the surface table.
* fix(connectors): the panel re-mints only a dead link
Try again on a failed or expired target mints a fresh link on the open operation. A waiting
target keeps the link it was minted with; the card reopens it and connectors.connect refuses
to spend a second mint (LINK_STILL_VALID). The unused refresh_link() goes. The package
docstring names the new siblings; the nine-name public surface is unchanged.
* test(connectors): the local-batch test answers the operation the way the card does
The callback stopped returning an answer in f782b26d98 (the card answers through
connection.respond); this test still returned one and waited out the 300s deadline in CI.
* ci: retrigger
* fix(connectors): the desktop card appears outside guided onboarding
Live on a signed-in macOS desktop, the two-app connect never showed a card. Three
defects, each hidden by a test that bound state the running app never binds.
The backend read the surface from HERMES_SESSION_PLATFORM only. The desktop and TUI
gateway bind it as HERMES_SESSION_SOURCE (_set_session_context), so session_platform()
was "" and managed connects took the off-desktop branch: links in the model's message,
no operation. session_platform() now reads platform, then source. The E2E test binds
through server._set_session_context instead of set_session_vars(platform="desktop").
The renderer routed manage_connections to the card only under isOnboardingEnabled(),
the HERMES_GUEST_ONBOARDING launch flag, in message-parts.tsx and the run splitter in
fallback.tsx. tool-render-class.ts had already dropped that gate in this PR; the two
routers had not. Both now route on the tool name alone.
ConnectorTool resolved the session owner by the runtime id. Owner routes, hints and
session rows are keyed by the stored id, so in registry topology the owner never
resolved and the card rendered null while the tool blocked. It now resolves by the
stored id, matching the PR1.5 card and every other owner lookup.
message-parts-connectors.test.tsx mounts the real Fallback router with the onboarding
flag off and distinct runtime/stored ids; red before each renderer fix, green after.
* style(connectors): shorter comments, no module mock in the card router test
The router test mocked isOnboardingEnabled to false; jsdom has no preload bridge, so the
real function already returns false. Comments that restated the code are cut to one line.
anti-slop: no net-new findings (25 touched files)
* fix(connectors): Connect on a waiting row opens the stored link
ConnectorCard derived the button's loading state from the phase label, so a managed row that
read "Finish connecting in your browser" (every row, since links are minted up front) had a
disabled Connect button. Nothing on the desktop could open the sign-in link; every managed
connect ended skipped, not_connected, or at the deadline.
The card now takes `busy` for "the action itself is running" and keeps `phase` as a label.
The MCP card passes its in-flight flag; the connector card passes the re-mint wait. Red before:
the Connect button on an initiated row rendered disabled and a click opened nothing.
* fix(connectors): a settled card stays dead; the card binds to its tool call only
A second connect for the same apps revived the finished card on the old tool row. The
connection.request payload carried no id, so the renderer fell back to matching rows by
connector names, and any row with those names qualified, settled or not.
The operation now records the model's tool_call_id and sends it in connection.request and in
the resume snapshot. The card binds to the tool row with that id and to nothing else; the
name-match fallback is deleted. A payload without the id is rejected by the store.
`reason` is removed from the tool: it was the only text the card ever showed from the model
and its absence forked a second tool part, since `reason` doubled as the row-correlation key
in tool-parts.ts. The card never needed it.
`connection.expire` is deleted from the contract and from _EXPIRING_REQUESTS: the card is
raised with _emit, not _block, so nothing has emitted it since the operation lifecycle landed.
Sid's rule of record: a resolved card is fully dead; no path brings it back.
* fix(connectors): the watch loop settles once, on time, and never raises into the result
Three findings from the live review, one loop.
Continue racing a finished sign-in: the loop ran the gateway read, then settled. A read that
returned `connected` for an already-settled or failed target raised IllegalTransition out of
the tool and the model got a generic error instead of the per-app outcomes. The read now skips
targets that are not live (pending, initiated) and skips a settled operation; the loop checks
`settled` after every read.
Settle reason as row text: `settle()` wrote `continue`/`deadline` into each unresolved target's
`detail`, and the card printed it in red. The reason stays on the operation only.
Stop and the deadline waited for the next tick: `/stop` sets a per-thread flag with no wake
hook, so the sleep is sliced at 250 ms and the flag and clock are read each slice. The clock is
also checked before each read, not only after.
Tests: a failed mint that later reads connected settles cleanly; Continue during a read keeps
the settled result; no reason in detail; an interrupt settles within the same second.
* fix(connectors): MCP setup off the desktop returns unavailable instead of blocking
run_mcp_operation treated a non-None connection_callback as "a card exists". Every tui_gateway
session has that callback, the Ink TUI included, so an MCP install from the terminal UI blocked
until the 300 s deadline while the docs promised `unavailable` with the terminal commands.
The MCP path now reads the session surface the same way the managed path does; the callback is
never the predicate. Test binds the surface to `tui` with the callback attached.
* fix(connectors): a failed Try again shows the failure, not the old dead link
The panel's re-mint ignored the gateway's per-app status and moved the row to `initiated` with
whatever link came back, `None` included, so a mint that failed again rendered as waiting on the
link that had already died.
One reader of a mint response now serves both the first mint and Try again
(`managed.mint`, with the actor as a parameter). A repeated failure keeps the row `failed`,
drops the link, and carries the vendor's new text through `operation.refresh`, which emits a
frame without a state change so the card redraws.
* fix(connectors): a forced reconnect waits for the new sign-in before it reports connected
`reconnect` with `force: true` is the account switch. The vendor keeps the old account active
while the new link waits, so the first list read after the mint said `connected` and the
operation settled at once: the new link was dropped and the model was told the switch was done.
A forced target is marked awaiting_new_attempt after the mint. The watcher ignores its row until
the list shows the new attempt (`connectionStatus: initiated`) once, then trusts `connected`.
* fix(connectors): the operation registers under the gateway session key
The tool registered the operation under the agent's session_id; every RPC (connection.respond,
connectors.operation.status, the panel's connectors.connect) and the update emitter looked it up
by the gateway's session key. Those agree until compaction rotates the agent id mid-turn; then
the card's clicks find nothing, no update reaches it, and the tool waits out the deadline.
The registration key is now the bound HERMES_SESSION_KEY, with the agent id as the fallback for
callers with no gateway (unit tests, a bare CLI). The E2E passes a rotated agent id and drives
the card by the gateway key.
* fix(connectors): the forced-reconnect gate reads any non-active row; a failed re-mint of an expired row is failed
Three follow-ups from the verification of the fix pass.
The awaiting_new_attempt gate cleared only on the literal `connectionStatus: initiated`. The
field is optional on the wire and `initializing`, `failed`, `expired` are valid values, so a
forced reconnect could wait the full 300 s and swallow a failed new attempt. The gate now holds
only while the row still reads as the old account (`connected` or `active`) and releases on
anything else.
Try again on an `expired` row whose re-mint fails raised IllegalTransition (no expired → failed
edge). The re-mint steps through `initiated` as the user's attempt, then `failed`, then drops the
dead link.
`detail` never carries a state name any more: `failed` as detail rendered as the row label and
made agent/display.py tag the settled result as a tool error. Only vendor text goes there.
`connection.expire` removed from the renderer's unscoped-stream set; nothing emits it.
* refactor(tools): discovery also scans tools/<pkg>/tool.py
A tool family that is a whole package had no way to register: discovery
globbed tools/*.py only and derived the module name from the filename.
Now the candidate list is tools/*.py plus tools/*/tool.py, merged and
sorted once so import order does not depend on depth (register() lets a
same-name duplicate overwrite silently), and the module name comes from
the path relative to tools/. Only tool.py is scanned inside a package,
so its siblings are libraries by construction. A package without an
__init__.py is skipped with a warning rather than registering from a
checkout and vanishing from the installed wheel.
The AST prefilter and the (mtime, size) disk cache are per absolute path
and work unchanged. The two hand-rolled tools/*.py enumerators in tests
now use the same candidate helper.
* refactor(connectors): one package for the connector domain, tools/connectors/
The connector code was spread across six flat files and a root-level
module that was a sibling of model_tools.py only by address:
tools/connections_tool.py -> tools/connectors/tool.py (schema, register, dispatcher)
tools/connectors/managed.py (the managed leg, split out)
tools/connections_tool_mcp.py -> tools/connectors/mcp.py (validation split out ->)
tools/connectors/targets.py (normalize_targets, validate_action)
tools/connections_tool_operation.py -> tools/connectors/operation.py
tools/connector_search.py -> tools/connectors/search.py
model_tools_connectors.py -> tools/connectors/dispatch.py
tools/tool_gateway/ -> tools/connectors/gateway/
Move only; every function body is unchanged. tools/connectors/__init__.py
is the door: nine names, the whole cross-package surface. model_tools and
tool_search deep-import a few helpers past it on purpose and the docstring
says so. The two split files make the import graph one-directional
(tool -> mcp -> targets, tool -> managed) where the old layout had
connections_tool importing validation out of the MCP file.
One behaviour-neutral seam change: the _connectors_available try/except
wrapper is gone. connectors_available() already fails closed, and both
the registry handler and the inline executor now read it as a module
attribute (gateway.config.connectors_available), so tests patch it in
one place instead of two. _default_client lives in managed.py, the only
module that calls it.
tools/managed_tool_gateway.py and tools/managed_gateway_auth.py stay:
they are gateway identity shared by tts, transcription, image and modal.
Test files follow their modules. No docs referenced the old paths; no
compat pointer is added (in-tree moves get none).
* ci: retrigger (zero-job dispatch on 142466de6b)
* feat(connections): manage_connections covers local MCP servers; setup_mcp leaves the schema
One model tool now connects the user to apps of both kinds. A target
`{"name": "linear", "mcp": true}` is a locally configured MCP server;
`install` / `enable` / `authorize` are its verbs. Bare strings and
`{"name": ...}` stay managed connectors and that leg is unchanged.
MCP targets run through one backend-owned connection operation
(tools/connections_tool_operation.py): created with a server-side
deadline from the new config key `connections.wait_timeout_seconds`
(default 120, floor 5, no ceiling), per-target state, and exactly-once
settlement (all resolved / Continue / deadline / interrupt). Unresolved
targets freeze as `not_connected` with the settle reason.
Why the fold works now: the approval card is reached through
`agent.connection_callback` via the agent-level inline executor table,
which is the only path that carries a GUI callback. Registry dispatch
(every non-GUI surface) settles MCP targets as `unavailable` with the
`hermes mcp install / login` hint; managed targets in the same call
are unaffected.
`setup_mcp` is removed from every advertised toolset and from the
deferral list; an inline-table shim keeps calls from conversations
opened before this change dispatching (prompt-cache protection).
`_LEGACY_TOOL_ALIASES` is not the mechanism: inline tools bypass it.
Gateway: `mcp.setup.request/respond` are replaced by
`connection.request/respond/expire` (no wire compat; desktop ships
with this). The bridge waits exactly the operation's deadline. The
`session.resume` snapshot gains `pending_connection` so a reopened
window restores the card with the original deadline.
`manage_connections` joins `_SEQUENTIAL_DEADLINE_EXEMPT_TOOLS`: the
operation owns its wait; the 420s guard must not report `tool_timeout`
while the card is live.
The portal `check_fn` on the tool is dropped in favour of a
handler-level gate on the managed leg, so signed-out sessions can still
approve local MCPs.
* wip(desktop): connection.request store, resume restore, card routing for MCP targets
Renderer half of the setup_mcp fold, first slice: connection-request store
(mirrors clarify), connection.request/expire handling, pending_connection
resume restore, mcpTargets() + isCardTool(name, args) so MCP-target
manage_connections calls classify as cards. Not yet: the card component
rewrite (mcp-setup-tool.tsx), mcp-directory.ts removal, vitest, docs.
Does not typecheck until the card rewrite lands.
* fix(config): hermes update turns on the connections toolset for saved toolset lists
`hermes tools` writes an explicit `platform_toolsets.<platform>` list, and the
resolver reads absence from that list as "unchecked". The `connections`
toolset (#106842) shipped after most users last saved, so `manage_connections`
is stripped from the schema on every install that ever opened the picker.
The Nous entitlement gate never runs; the agent reports the tool as missing.
Migration 44 -> 45 (renumbered when folded into #109517; main was already at 44) appends `connections` to each explicit per-platform list
that lacks it and records the offer in `known_builtin_toolsets` where that
record exists, so a later uncheck reads as a decline. It skips: platforms
whose record already holds `connections` (the user saw the checkbox and left
it off), bare composite lists ([hermes-cli]) that already inherit it, platforms
where the toolset is not allowed, and any config whose `agent.disabled_toolsets`
names `connections` (Blank Slate, `hermes tools --disable`), because the
resolver subtracts that list last and the enable would never take effect.
The explicit-list test is the resolver's own: any configurable or plugin key.
`hermes update` runs migrations post-pull for the active profile and every
sibling, so one update is enough. Fresh installs and composite users were
never affected.
* refactor: anti-slop pass on the desktop slice; shorten added comments
Parse connection.request at the boundary with a typed wire interface instead of
unknown + typeof; mcpTargets reuses connectorText; comments cut to one or two
lines. slop-ratchet: no net-new findings in 13 touched files.
* feat(desktop): the MCP approval card answers manage_connections; MCP Directory removed
The existing card (mcp-setup-tool.tsx) now reads the connection-request store,
renders for manage_connections calls with mcp:true targets, answers through
connection.respond with a per-target outcome, and no longer calls reload.mcp
after Install; the new server's tools arrive on the between-turns refresh.
A settled operation renders the first target's frozen state.
session.resume restores a pending card with its original deadline on both the
activate and cold-resume paths.
lib/mcp-directory.ts is deleted along with its two fallback branches
(suggestion provider, card install). The catalog was already primary in both;
a catalog miss now yields no suggestion / a notInCatalog error. The GitHub
never-suggest test is rewritten on catalog-shaped data.
vitest: connection-request store (6), suggestion provider, clarify restore.
slop-ratchet: no net-new findings in 19 touched files.
* chore: drop __pycache__ files swept in by an over-broad git add
* fix(desktop): correlate the connection.request row with the model's tool call by reason
The synthetic row from connection.request and the tool.start row carried
different ids and no shared match value (op_id is not in the model's args),
so the card mounted twice. reason is the arg both sides carry.
* docs: manage_connections covers local MCP servers; connections.wait_timeout_seconds
* fix(connections): settle reason derives from target state, never from the renderer
A card that answers one of two targets and claims all_resolved must settle as
continue with the other target not_connected; found live with a two-target call.
* fix(desktop): a pending connection card re-arms on resume and activate
The store entry was restored but the transcript row was not, so navigating
away and back (or reloading) lost the card while the backend kept waiting.
restorePendingClarifyToolCall's core is generalized to any blocking tool
name and both resume paths project the connection row through it.
Verified live: card restored after navigate-away and after a full renderer
reload, deadline_at unchanged, approve settles connected.
* style: literal wording in added comments, docstrings and docs
* fix: shared gateway-event contract and config-schema category for the connection events
connection.request/expire replace mcp.setup.* in apps/shared gateway-events
(json list, BACKEND_EVENT_NAMES, GatewayEventMap) so the renderer's event
union includes them and the tui_gateway contract test passes. The new
`connections` config section folds into the agent tab like the other
single-field sections.
* style: import order (perfectionist) in the desktop and shared files this PR touches
* chore: retrigger CI (zero-job dispatch failure, auto-heal)
The summary path had grown a verbatim copy of turn_api_request's 3-line
surrogate/ASCII chokepoint — the same drift class this PR removes for the
hand-rolled kwargs builder. Move the two lines and the #50959 rationale into
`message_sanitization.sanitize_outbound_kwargs` and call it from both sites,
so the next sanitizer step added to the main loop cannot miss the summary.
Tighten two comments: "same kwargs builder" (cache_control redecoration is
not re-applied here) and a `_summary_text` note that is true for all three
summary branches, not just the chat one.
`_summary_text_with_scrub` was the old helper plus a warning — `.content`
never carried tool calls, so nothing was being discarded that was not
already ignored. Restore the original name, keep the warning with the WHY
(the request now carries tools, this path never executes a call), and make
the tool-only test assert the log line so it actually binds the new code
path instead of passing on the pre-fix retry behaviour. Trim the builder
comment to the WHY.
Building the summary through `_build_api_kwargs` means it now carries
`tools`, but it still skipped the `_sanitize_structure_surrogates` /
`_force_ascii_payload` chokepoint the main loop applies after building
(turn_api_request). On cache-planned routes the main loop scrubs a deep
copy of the tools, so `agent.tools` can still hold a lone surrogate that
providers reject with a non-retryable 400 (#50959 class) — a request the
tool-less summary never used to make. Apply the same two passes here.
`_chat_summary_attempt` now builds through `_build_api_kwargs`, leaving
`_iteration_summary_chat_kwargs` (56 lines mirroring the transport by hand)
and its only consumer `AIAgent._resolve_lmstudio_summary_reasoning_effort`
without a caller. The transport already owns every quirk they re-derived
(fixed temperature, LM Studio `reasoning_effort`, portal tags, provider
preferences, pareto router plugin), so there is nothing to keep in sync.
Three inline context-manager classes shared the same __enter__/__exit__
boilerplate and differed only in the events yielded before the raise.
Also drop the unreachable 'or agent.base_url' fallback: every
anthropic_messages init path sets _anthropic_base_url, and the two
sibling call sites read it bare.
The SDK's malformed-frame ValueError substrings were hand-copied into the
retry classifier and the error summary; a third SDK message would need
both remembered. The mid-tool retry branch no longer clears
partial_tool_names itself — _start_stream_attempt resets it for every
attempt. buffer_anthropic_tool_input's docstring states the trade: the
knob stays on the turn's kwargs and the changed tools block costs one
prompt-cache miss.
A tool_use block name recorded by a stream attempt that died before any
visible text survived into the next attempt: only the deltas_were_sent
mid-tool branch cleared result["partial_tool_names"]. When the retry then
streamed plain text and dropped, the partial stub blamed the stale tool
("Stream stalled mid tool-call (old_tool)") and the stale name could make
a later attempt look mid-tool-call when deciding whether the drop is
retryable. Reset it in _start_stream_attempt alongside
provider_tool_in_flight, which already has attempt-local semantics.
Regression: two-attempt stream (tool_use start + parse error, then text +
drop) — stub content and emitted deltas carry no stale tool name.
Follow-up to the cherry-picked fix: keep the classifier widening
("expected value at line" is now a transient stream parse error on the
main turn) but replace the messages.create() fallback with a retry on the
same stream wire.
Why not create(): the fallback ran outside Relay (lost request rewrites),
outside _handle_stream_error (could replace text already shown to the
user with a different generation), ticked no liveness events for the
whole buffered payload, and a bare identical retry still re-emits the same
malformed JSON.
Why not drop the fine-grained-tool-streaming beta (#108583/#109056): live
probe on claude-sonnet-4-5, ~500-line tool call - beta on: max inter-event
gap 1.6 s; beta off: 139 s zero-event gap while Anthropic buffers the
args, which the 180/240 s stale-stream detector kills on larger payloads
(the regression 80a899a8e2 fixed).
Instead, on a parse error the retry sets `eager_input_streaming: false`
on every tool for that request only (the SDK/API per-tool field overrides
the legacy beta header), so Anthropic returns buffered, server-validated
args while the happy path keeps fine-grained streaming. A tool_use that
started streaming is registered in partial_tool_names so the mid-tool
transient retry fires the same way it does on the chat_completions wire.
The "screen unchanged" result points the model at its previous capture. After
context compression that capture may be summarized away, so the note would refer
to pixels no longer in context. Mirror read_file's reset_file_dedup: the
compaction boundary (both the summary path and the codex app-server path) now
clears the session's screenshot digest, and the first capture afterwards delivers
the image again even when the screen is byte-identical.
Trimmed salvage of #110045 (deltas 1 + 2 only), stacked on the #110009 scope inheritance:
- `declared_conversation_scope` treats an inherited value as a DECLARED scope only when it
carries the `gwk_` prefix. A rotated CLI parent publishes no affinity scope (None → sticky
key falls back to the conversation root); the fork now publishes exactly the same instead
of an explicit physical lineage root. `resolve_prompt_cache_scope` honors any inherited
value directly, so the body `prompt_cache_key` still matches.
- `build_cache_parity_fork` snapshots `parent._conversation_root_id()` as
`_cached_conversation_root`; with `_session_db=None` the fork's own walk fell back to the
parent's PHYSICAL id, so after a compression rotation the review's Portal
`conversation=` tag fragmented usage attribution across one logical conversation.
Dropped from the original: copying `_gateway_session_key` onto the persistence-detached
fork (no cache-identity consumer reads it there; the compression-boundary hooks were
deliberately severed by `_detach_fork_compression`), and the defensive
hasattr/callable/try wrapper around `_conversation_root_id()`.
build_cache_parity_fork gives the same-model fork the parent's session_id,
cached system prompt, tools[] and session_start — but with
_persist_disabled=True and _session_db=None, BOTH cache-identity resolvers
diverged from the parent on their own: declared_conversation_scope failed
closed on _persist_disabled, and the lineage walk skipped on the missing
DB. The fork's affinity header (set_affinity_scope) and body
prompt_cache_key (cache_scope_id on the OpenAI-wire transports) therefore
keyed a different bucket than the gateway parent, costing one cold
~full-context request per review. Not gateway-only: any parent whose
lineage root != current physical id diverges too (teknium1's triage table).
Fix, per the triage's suggested direction: on the not-routed branch only,
the fork stamps _inherited_cache_scope = resolve_prompt_cache_scope_safe
(parent) — the parent's ALREADY-RESOLVED scope, no DB access from the fork,
persistence fully detached. Both declared_conversation_scope and
resolve_prompt_cache_scope return the inherited scope first when set, so
the header path and the body path are fixed together (fixing only one
leaves the other divergent — Vivamisu's header/body split observation).
Routed (different-model) forks, /branch children, delegate/tool children
and fresh sessions set nothing; the fail-closed default stands untouched.
/btw shares build_cache_parity_fork and gets the repair for free.
Clean-room port of the approach in zed-industries/zed#63342. The native
Gemini adapter previously down-translated every tool schema into the
restricted FunctionDeclaration.parameters subset, which was lossy: anyOf
unions without an outer type, bare arrays, $ref/$defs indirection and
additionalProperties had to be stripped or repaired, and one
unrepresentable construct could 400 the entire request (live repro:
INVALID_ARGUMENT ...properties[bare_array].items: missing field).
Google now accepts plain JSON Schema in parametersJsonSchema on all
current models. The adapter sends full schemas through that field; the
old subset translator is replaced by a light normalizer that deep-copies,
strips root $schema, inlines same-document $refs (MCP pydantic / zod
emit them; unresolvable or circular refs pass through untouched with the
reason logged), and guarantees an object root.
Live-verified against the real API: the union+bare-array+$ref schema
that 400s through the legacy parameters field is accepted with 200 via
parametersJsonSchema on gemini-3.7-flash and gemini-2.5-flash, and
gemini-2.5-flash returns a correct functionCall against it.
The review fork is a full AIAgent, not an auxiliary_client call, and its
routed branch deliberately skips the parent's reasoning_config (the parent's
effort vocabulary may be invalid for the routed provider). It also never read
the per-task key, so an explicit `auxiliary.background_review.reasoning_effort`
was silently ignored and the routed fork ran at the provider default (#94825).
Routed forks now parse the task key through the shared parse_reasoning_effort
(same levels and `none` alias as every other aux task); unset keeps the
provider default, an unknown level warns and falls through. The same-model
path is untouched: it still inherits the parent's reasoning_config verbatim
for prompt-cache parity.
Salvaged from #94832 (liuhao1024), re-applied on the decomposed
_fork_init_kwargs seam.
Co-authored-by: liuhao1024 <sunsky.lau@gmail.com>
auxiliary.background_review.reasoning_effort was a silent no-op on the
same-model path: the fork inherits the parent's reasoning_config verbatim to
keep prompt-cache parity (#30532), and nothing told the user. Emit a one-time
user-visible warning (parent-scoped, so a nudge-per-turn session warns once,
not per fork) when the key is actually set, document the no-op in the config
reference, and leave the fork-birth request bytes unchanged.
The Python-repr pass ran after the MCP probe header scrub and collapsed
'Authorization': 'Digest ***' to '***', erasing the scheme word that scrub
deliberately keeps (tests/hermes_cli/test_mcp_probe_redaction.py went red
on rebase). Skip values that already carry a mask marker.
Port from OpenHands/software-agent-sdk#4508: their dict-entry secret
redaction was uppercase-only and leaked mixed-case keys (UserPassword,
sessionToken). Apply the same case-insensitive treatment to the Python
mapping-repr pass: a casefolded credential suffix (apikey/token/secret/
password/passwd/credential) now qualifies a key, while metadata names
(TOKEN_COUNT, password_policy, tokenizer) stay untouched.
The streaming accumulator dropped delta.reasoning_details entirely — only
non-streaming responses preserved the OpenRouter unified reasoning replay
data (signatures, encrypted reasoning blocks). Providers that require the
reasoning_details sequence passed back on the next turn (OpenRouter
reasoning models, Anthropic signed thinking via shims) lost continuity on
every streamed turn, which is nearly all turns.
- Accumulate delta.reasoning_details (attr or model_extra) during
streaming and attach the merged list to the final mock message, where
_build_assistant_message's existing passthrough persists it.
- _append_streamed_reasoning_detail merges consecutive reasoning.text /
reasoning.summary fragments into one logical entry (OpenRouter streams
them as word-level deltas; unmerged they bloat the replayed signature
payload) while encrypted/opaque entries stay discrete. Later fragments
backfill signature/id/format/index fields the first fragment omitted.
Ported from earendil-works/pi#8605 (commit c5ad7c1b0), credit
@cristinaponcela for the merge-fragments pattern.
Tests: tests/run_agent/test_streamed_reasoning_details.py (8 tests;
sabotage-verified: disabling accumulation fails the E2E test, disabling
merging fails 4 unit tests). Neighboring suites green (test_streaming.py
40 passed).
Small title models parroting a prompt example back verbatim produced
sessions named "Fix login button on mobile" with no relation to the
conversation. The example lines in _TITLE_PROMPT_TEMPLATE now render
from _PROMPT_GOOD_EXAMPLES so the guard set and prompt cannot drift,
and generate_title rejects exact (case-insensitive, wrapper-stripped)
echoes so the instant derived title survives instead. 'Friendly
greeting' stays allowed — it is prescribed output for bare greetings.
The ChatGPT Codex models endpoint interprets client_version as a Codex
CLI compatibility version and filters out any model whose
minimal_client_version is newer than the value sent. Hermes hardcoded
client_version=1.0.0 at both catalog request sites, so model visibility
was accidentally coupled to a version scheme Hermes doesn't follow —
future models gated behind a higher minimal version would silently
vanish from the account catalog.
The backend accepts the exact sentinel 0.0.0 as an ungated request
returning the complete account catalog (verified live: 0.0.0 and
current versions return identical model sets today, while omitting the
parameter is HTTP 400 and out-of-sequence values like 0.0.1 return no
models). Both request sites (hermes_cli/codex_models.py and the
context-length probe in agent/model_metadata.py) now share one
CODEX_UNGATED_CLIENT_VERSION constant.
Clean-room port of the observed behavior in zed-industries/zed#62729;
no GPL code translated.
Clean-room port of the billing-code coverage from zed-industries/zed#63208: credit_balance_exhausted, organization_spend_limit_exceeded, project_spend_limit_exceeded, organization_usage_limit_exceeded now classify as billing (rotate + fallback) instead of falling through to generic buckets.
Some OpenAI-compatible gateways fronting Gemini backends emit the native
uppercase finish reasons (STOP, MAX_TOKENS) instead of the lowercase
OpenAI contract values. Every downstream comparison in Hermes uses
lowercase literals, so an uppercase reason silently fell through: a
clean STOP completion missed the stop handling and a MAX_TOKENS
truncation never entered the length-recovery path.
Adds normalize_finish_reason() as the single owner in
agent/message_sanitization.py (case fold + alias map: max_tokens->length,
end->stop, function_call->tool_calls) and wires it at both wire-intake
choke points: ChatCompletionsTransport.normalize_response and the
streaming chunk-capture loop in chat_completion_helpers. Non-string and
empty values pass through unchanged so existing 'or "stop"' defaults
and the Poolside int-reason path keep their behavior.
Adding grep/awk/sed to _FILE_READ_COMMANDS made the PATTERN operand
participate in the secret-file predicate, so `grep .bashrc app.py` or
`grep -n .env src/settings.py` — reads of SOURCE files — ran the
ENV/YAML assignment pass and masked opaque values that main leaves
alone. Skip the first non-flag positional for the pattern-first
readers, as #109369 originally did, so only real file operands gate.
`cat $HOME/.hermes/config.yaml` was ungated because the `$` bail-out
fired before the `.hermes` segment was inspected; strip `$HOME/` and
`${HOME}/` like the HERMES_HOME prefixes.
Review follow-up on #110228.
Fold KoNit-K's `_command_reads_secret_bearing_file` and the pre-existing
`_command_reads_env_file` into a single `_command_reads_secret_file` so the
`code_file` gate in `redact_terminal_output` has one owner: `.env`-style
basenames and shell rc/profile files anywhere, `config.yaml` only under a
`.hermes` directory or `$HERMES_HOME` (arbitrary project YAML stays on the
code_file path). `grep`/`awk`/`sed` join the reader set instead of a second
table with a positional-argument special case: on a file read, any non-flag
operand that names a secret-bearing file is enough — the pattern/program
operand never matches a basename, so the extra rule bought nothing.
Tests: the negative parametrization now uses an opaque credential-shaped value
(the placeholder it used before would never have been masked on either path,
so the "stays unredacted" half proved nothing) and asserts the same value IS
masked under `cat .env` in the same test.
Keep the narrow basename allowlist, but do not treat $HERMES_HOME as an
unresolved path, and split pipelines only on unquoted |;&.
Co-authored-by: Cursor <cursoragent@cursor.com>
The text-only drop guard in _finish_chat_stream required content_parts,
so a stream that died while still emitting delta.reasoning (no
finish_reason, no usage) fell through to the synthesized "stop". With
the reasoning-only clean-stop promotion in finish_text_response that
stamped "stop" turned the truncated thought into the final answer,
where main entered the continuation ladder. Extend the guard with
reasoning_parts so the drop yields the partial-stream stub and the
ladder still runs; a real clean stop carries finish_reason="stop" and
is unaffected.
Review follow-up on #110227.
Follow-up to KoNit-K's commit: rebuild the promotion on the existing
`agent._extract_reasoning` helper (the same reader the ladder terminal and
`build_assistant_message` use) and write the promoted text back onto
`assistant_message.content` so the persisted assistant row carries the answer
as ordinary content. Without that the transcript tail was an assistant row with
empty content and only `reasoning`, which `drop_thinking_only_and_merge_users`
strips from the next request — the model would see its own answer vanish on a
"continue" turn.
Tests: trim to the two invariants (clean stop → one API call, persisted as
content; `finish_reason == "length"` → never promoted, continuation still
owns it) and keep the truly-empty terminal case. The prefill wire-payload
regression test now drives a non-clean-stop reasoning-only reply, which is the
only shape that still reaches the prefill rung.
`_store_cached_client()` refuses an `_AuxProbeClientStub`, but
`_get_cached_client()` assigns to `_client_cache` directly and so never
reaches that guard. check_fns resolve through this path inside
`aux_probe_mode()` during tool-schema assembly, and the cache key carries no
probe/runtime distinction — so the stored stub is returned to the next real
caller sharing that key, which dies on attribute access with
`_AuxProbeClientStub used as a real client (attribute 'chat')`.
The `async_mode` field in the cache key is what kept this latent: the probe
caches the sync variant, so async consumers (`analyze_image`) miss the entry
and build a real client, while sync consumers (`browser_vision`) hit the
poisoned one and fail on every call.
Observed against a local OpenAI-compatible vision endpoint, where
`browser_vision` failed every call with that RuntimeError while
`analyze_image` against the same provider worked.
Guard the inline store the same way `_store_cached_client()` does, and
return the stub to the probe caller without caching it.
The existing `test_probe_stub_never_cached` pins the invariant only on
`_store_cached_client()`, which is why the unguarded door went unnoticed;
the new test exercises `_get_cached_client()` and fails without this change.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`scripts/run_tests.sh tests/<dir>/` is how a change gets its regression
coverage run, so a test filed under the wrong directory is a test nobody
runs when that code changes. Two kinds of drift had accumulated.
Parallel directories for one source package, folded into the mirror:
tests/acp -> tests/acp_adapter (its __init__/conftest move with it)
tests/cli -> tests/hermes_cli (prompt_toolkit fixture merged into
hermes_cli/conftest.py)
tests/run_agent -> tests/agent (backoff fixture becomes
agent/conftest.py)
tests/relay -> tests/gateway/relay
tests/state -> tests/hermes_state
246 loose files at tests/ root, routed by the package they import/patch:
hermes_cli, hermes_state, agent, gateway, tools, plugins, tui_gateway, cron.
Installer and desktop-update script tests go to tests/scripts/{install,
desktop_update}/. 43 tests of root-level modules (batch_runner, utils,
hermes_constants, packaging) stay at the root.
Filenames drop their issue numbers (95 files: test_89315_x.py -> test_x.py);
the number stays in the module docstring where it has context.
Collisions: test_cli_skin_integration.py existed in both tests/ and tests/cli
with different subsets — merged into one (10 tests, all kept);
run_agent/test_pre_compress_memory_context.py -> agent/..._handoff.py;
tests/test_account_usage.py -> agent/test_account_usage_fetch.py;
tests/test_web_server.py -> hermes_cli/test_web_server_ws_ping.py.
Deleted: test_minisweagent_path.py (empty since PR #2804),
test_model_picker_scroll.py (tested a private copy of the logic, imported
nothing), test_process_loop_event_loop_warning.py (asserted asyncio behaviour,
imported nothing from Hermes).
Repo-root path arithmetic (Path(__file__).parents[N], dirname chains) is
bumped for the 202 files that changed depth and verified by evaluating every
such expression against the new location. classify_changes' desktop-updater
lane prefix, tests-os.yml's ignore glob and every in-tree path comment follow
the moves. tests/test_tests_tree_layout.py keeps the tree from drifting back.
resolve_provider_client("commandcode-anthropic") with no explicit api_mode (a bare
``auxiliary.<task>.provider`` entry) built a plain OpenAI client: _wrap_transport
only consulted req.api_mode and URL heuristics, and api.commandcode.ai/provider/v1
matches none. With _reasoning_config now emitted for that profile, an unwrapped
client would TypeError on the unknown kwarg; before, the OpenAI-wire call simply
misfired against a Messages endpoint. Fall back to the registered profile's
api_mode so the wrap and the reasoning gate agree.
commandcode-anthropic (api_mode=anthropic_messages, OpenAI-shaped base URL) lost
thinking control on compression/title/vision calls after #109530: once its class
overrides build_api_kwargs_extras, _project_provider_profile marks the profile as
handling reasoning and drops the generic extra_body.reasoning fallback that the
Anthropic adapter used to read — and the _reasoning_config gate only fired for
provider=anthropic, Portal /v1/messages ids, /anthropic URLs and MiniMax.
Carry the profile's api_mode on the projection and include it in that gate, so
any anthropic_messages profile reaches the adapter regardless of URL shape.
Chat-completions profiles are unchanged.
Before: _build_call_kwargs("commandcode-anthropic", claude-haiku, {"enabled": False})
-> no _reasoning_config, no extra_body.reasoning; adapter defaults thinking.
After: -> _reasoning_config={"enabled": False}.
Gate findings on the #108614 salvage (2c + simplify-code):
- CopilotACPProfile.fetch_models promised None on failure but raised (AuthError on a
missing CLI, RuntimeError/TimeoutError from the probe) — only the caller in
hermes_cli/models.py caught it; any other caller got exceptions. Wrap the body and
return None so the docstring and the base ProviderProfile.fetch_models contract hold.
- timeout_seconds was a PER-REQUEST budget (initialize + session/new each got the full
timeout => ~30s worst-case foreground stall on a hung CLI). One shared session deadline.
- /model --refresh (clear_provider_models_cache) wiped the disk cache but not the new
session memo, so a fresh CLI login stayed stale for 5 min past an explicit refresh.
- A failed probe is now memoized for 30s instead of the full 5 min, so signing in to
the CLI is picked up on the next switch.
- _fresh_acp_memo fixture restores the memo on teardown — no cross-test state leak.
- build_api_messages reads agent._current_turn_timestamp directly: a caller that skipped the
turn prologue now raises instead of silently falling back to per-request wall time, which
would re-create the mid-turn drift the fix removes. Only production caller
(assemble_api_request) runs after _reset_per_turn_agent_state; cross-reference to the
tripwire _inflight_turn_started so the two clocks are not "unified" by mistake.
- is_dangerous_confirmation lowercases once instead of once per pattern (now on the per-request path).
- Tests: one _send(idx=) helper instead of three spellings of the builder call; the
untrustworthy-stamp contract is its own test.
Two replay-boundary findings from the #109320 review (gaoanze888, andrexibiza):
- is_interrupted_tool_result matched the bracketed marker anywhere in the payload, so a
successful `read_file`/`search_files` result QUOTING "[Command interrupted]" (a doc, a
grep hit) was classified as a killed run — the read-only block vanished from the
model-facing history, a `terminal` result became the UNKNOWN-effect notice — while
send/replay parity still held because both consumers made the same mistake. Require
the shape every executor actually produces: the marker is the LAST line of `output`,
inside a JSON envelope with a non-zero exit code (or a bare text result). Real
interruptions from tools/environments (130), managed_modal (130) and
code_execution_tool (-1) all keep matching.
- strip_stale_dangerous_confirmations only bounded age from above, so a finite FUTURE
stamp (clock skew, corruption) had negative age and kept the confirmation plus its
api_content sidecar live. Freshness is now 0 <= admission - ts <= expiry; a future
stamp expires like a corrupt one. Missing stamps (legacy rows) stay untouched.
Test: the parity fixture carries both quoted-marker results (JSON grep hit, bare doc text)
and a genuine execute_code interruption envelope; the expiry test covers "nan" and two
future epochs. Red under: marker-anywhere, any-line, old loose heuristic, upper-bound-only.
Follow-up to the two cherry-picked commits from #105308 (@JoaoMarcos44), closing the
three blockers raised on that thread plus one regression the salvage found:
- Prefix-only on the send path. build_api_messages now canonicalizes only
messages[:current_turn_user_idx]; rows the current turn appended (its own tool
calls/results) pass through verbatim. Canonicalizing the live tail rewrote a block
the previous iteration had already sent whenever a tool result matched the
interrupt heuristic, which is exactly the mid-turn prefix rewrite this fix exists
to remove, and it also made the dangling-tail transform order-dependent on when
the user row was appended.
- Exact interrupt marker. is_interrupted_tool_result matched
"exit_code" + ("130" | "-1") + "interrupt" as substrings, so an ordinary
`grep KeyboardInterrupt` result next to a diff hunk header rewrote a terminal
result to an orphan notice (or dropped a read-only block). That heuristic was
tolerable at resume time only; it now runs per request. Match the executors'
bracketed markers ("[Command interrupted", "[execution interrupted") and nothing else.
- Admission-time clock. The frozen expiry clock was the input's platform-event
stamp, so a message queued 70 s before the turn ran kept a 129 s-old confirmation
live on the send path while replay expired it. _reset_per_turn_agent_state stamps
time.time() once at admission; the three other writes (bind identity, stage
message, build_api_messages write-back under suppress(Exception)) are gone.
- Fail closed on corrupt stamps. A present-but-unparseable timestamp (`"nan"`,
`"not_a_number"`) made strip_stale_dangerous_confirmations keep the confirmation
and its api_content sidecar. Coerce through hermes_cli.timefmt.coerce_epoch and
treat an unknowable age as expired; missing stamps (legacy rows) are still left
alone.
- Shape: drop the canonicalize_history_for_send alias (no consumer, never existed on
main), the `now=` kwarg (no production caller), and the getattr/hasattr rewrite of
_reset_per_turn_agent_state (only the test double needed it).
- Tests: 17 → 2 invariant tests. Real SessionDB round trip → canonicalize →
ChatCompletionsTransport bytes, equal to the send path with sidecars applied and
the durable list untouched, live tail preserved; admission-clock freeze across
iterations + corrupt-stamp fail-closed. Each is red under the matching mutation
(send path unpatched, whole-list canonicalization, per-request clock, fail-open,
loose heuristic).
Also widen the two real-subprocess list_models tests from a 2 s to a 30 s response deadline:
spawning a Python interpreter under 16 parallel test workers occasionally exceeded 2 s and the
file flaked (TimeoutError in _request). The deadline only bounds a failure; the happy path
returns as soon as the fake server answers.
docs/ was not the documentation site; it was a grab bag of long-form
design notes, wire contracts and observability guides that landed with
feature PRs because their authors needed somewhere to put them. Root
AGENTS.md already says long-form dev docs live in
website/docs/developer-guide/; this moves the 14 living documents there
(or to the matching user-guide section) so they are published, searchable
and linked from the sidebar instead of being found by grep only.
Developer guide: micro-compaction, gateway-session-lifecycle (was
session-lifecycle), state-db-recovery, multiplexing-gateway,
chronos-managed-cron-contract, relay-connector-contract, observer-hooks
(was observability/README), gateway-monitoring (observability/monitoring),
relay-shared-metrics, middleware, streaming-tts, billing-lifecycle.
User guide: egress/network-isolation (was security/network-egress-
isolation), features/kanban-multi-gateway (was kanban/multi-gateway).
Each page got title/description frontmatter and a sidebar entry; repo-
relative links became site links or GitHub blob URLs; two MDX brace
hazards escaped. Every in-tree pointer (module docstrings, config
comments, the relay conformance test's Path, the monitoring-doc test,
gateway-internals, cron-internals, kanban docs, .dockerignore, AGENTS.md)
now names the new location. `docusaurus build` passes with no unresolved
links on the moved pages.
