A delegation.model-only pin (provider inherited) took the unpinned
column because pinned was keyed on override_provider alone, so the
child inherited the parent chain and a mid-run failure could silently
swap the pinned model. model at the call site is creds["model"] --
delegation.model config -- at both call sites, so keying on it adds no
false pins. base_url-only pins already resolve to a provider override.
Caught by review on the PR; two matrix tests added (model-only pin with
absent and declared chains).
Adopted from PR #80421 with the author's explicit go-ahead on #80450
('Please proceed!'): config_defaults entry, cli-config.yaml.example
block, and user-guide docs for the delegation-scoped fallback chain.
Co-authored-by: Andrex Ibiza, MBA <84248988+andrexibiza@users.noreply.github.com>
Composes #80465's pin semantics with #80438/#80421's
delegation.fallback_providers semantics (#65038) and settles the
composition cell the three PRs leave undefined (#80450 map).
_resolve_child_fallback_chain implements the whole decision table in
one pure function: pinned children get no chain unless the delegation
section declares one (a delegation-scoped chain honors both explicit
intents; the silent drag in #80450 is specifically the PARENT chain
substituting a pin); an explicit empty list disables fallback; absent
config preserves historical parent inheritance exactly. Malformed
declared values log and fall back PIN-AWARE — never the parent chain on
a pinned child, so the config-error path cannot reintroduce the drag.
Entries normalize through the canonical get_fallback_chain.
The two existing inheritance tests are made hermetic against the real
user config (as #80438 also did).
Stress run with a real orchestrator subagent showed the per-task keys buried in results[] were not relayed
upward; the same prose lines now sit at the top of the sync delegate_task result. evals/subagent_process_handoff/
stress_handoff_live.py runs 12 real parent+child scenarios (handoff, orphan, unread, clean read, cap, exited refusal,
mixed fan-out, parent controlling the inherited process, sibling theft incl. adversarial, nested orchestrator,
5-way burst) and scores them against runtime accounting, never model prose.
A process that finishes while the child is alive needs no handoff, but if the child never
polls/waits/logs it, the result vanished: the completion notice is suppressed in the parent
and the child's summary never mentions it. Finalization now attaches exit code + output
tail as unread_completions, rendered in the parent's delegation notice.
The schema-diet test froze the verb list (a change-detector); it now asserts the enum
matches _SESSION_ACTIONS plus the by-name verbs. The cancellation harness gave a worker
thread 1 s to reach its transport, which a loaded CI runner missed twice this week; the
cancellation-latency assertion is unchanged, only the start-up wait is wider.
A child's background processes are killed at its teardown and their
notify_on_complete notices are suppressed in the parent, yet the child's
terminal result still said `notify_on_complete: true` and the parent's
delegation notice said nothing about processes left behind. Orchestrators
believed "CI watcher running" and waited on a completion that could never
arrive (recurring in the Sep 7 campaign sessions).
- process_manage(action="handoff", session_id, data="<purpose>"), children
only: process_registry.transfer_ownership flips owner_task_id/task_id/
session_key to the parent under the registry lock, so the completion is
stamped with the parent's owner at exit, passes the parent's sa- filter,
and is reaped by the parent, not the child. Cap 3 per child; an exited,
foreign, or non-child request is a tool error. The purpose rides the
event as handoff_note and renders in the parent's notice.
- Child terminal(background=True, notify=True) now returns
notify_on_complete=false plus a note: wait, kill, or hand off.
- _ChildRun.account_background_processes records handed_off_processes and
orphaned_processes on the result before cleanup kills the leftovers; the
parent's delegation block renders both.
`_on_tool_progress` bailed on the tool-progress gate before dispatching
`subagent.*`, so a Desktop/TUI user who hid tool-call chrome also lost the
subagent rows in the status stack and spawn tree. Subagent lifecycle is
application state (like `todo.updated`, clarify and MCP consent cards,
which already bypass the gate); the gate now applies only to the optional
progress chrome (reasoning previews, MoA rows, tool.generating).
The core-tool rename shipped `todo_list` on the wire (legacy alias `todo`
kept for old transcripts), but the Desktop renderer still matched the tool
by the literal `todo` in seven places: the live tool.start/tool.complete
mirror into the composer status stack, the todo-stream router, args
carry-over, the transcript hoist, the silent-tool class, the count noun,
and stored-history hydration. Every live task update therefore went into
the transcript as an ordinary tool row while the task panel stayed empty,
and reopening a chat never restored a finished list.
One predicate (`isTodoToolName`) now owns the wire/legacy name pair and
every site reads it.
Five test files and five evals scripts merged since this branch's base still import the
moved names via gateway.platforms.base. They work (base.py imports the names for its own
use) but the PR's invariant is that in-tree code imports from the defining module.
Same cleanup as the previous commit, applied to the third file that carried the
TYPE_CHECKING + in-function import pair. gateway.run imports none of the telegram/discord/
slack modules the conftest stubs, so import order is not a concern. test_feishu.py keeps its
TYPE_CHECKING import on purpose (FeishuAdapter is gated on optional lark_oapi).
Regression guard for the one-line import fix in commit 1; identical to the test in #102952,
whose /btw half is superseded by this PR. Red on origin/main (NameError), green here.
Self-review follow-ups on the F821 sweep:
- plugins/platforms/sms/adapter.py: the optional-import block now sets AIOHTTP_AVAILABLE like
the homeassistant / webhook / whatsapp_cloud adapters, and both call sites test the flag.
Removes the `if not aiohttp is not None:` double negation left by inlining
`_aiohttp_available()`.
- tools/mcp_tool_sampling.py: `call_context` defaults to `lambda: None` so the use site is a
single call instead of an Optional guard; the only None caller was a test. The
`from __future__ import annotations` was noise (`Context` is a runtime import). Comment
names the actual cycle (mcp_tool_server_run imports this module).
- gateway/platforms/helpers.py: drop the `from __future__ import annotations` — the only
MessageEvent annotations are attribute-target locals, which are never evaluated.
- tests/gateway/test_telegram_audio_vs_voice.py, test_video_context_note.py: module-level
`from gateway.run import GatewayRunner` like the ~100 sibling files; the
TYPE_CHECKING block + `# type: ignore[name-defined]` were contradicting each other.
(tests/e2e/conftest.py and test_feishu.py keep TYPE_CHECKING deliberately: they stub
telegram/discord before importing, and FeishuAdapter is gated on optional lark_oapi.)
Mutation check: neutralising the thunk read (`captured = None`) fails
test_captured_context_is_replayed_in_consent_call; restored → 14/14 green. ty on the three
touched production files vs origin/main: 0 new, 6 resolved.
Breaks the two import cycles that forced Protocol stand-ins in the F821 sweep, so the two
sites now name the real types.
gateway/platforms/event.py (new leaf): MessageType, ProcessingOutcome, MessageEvent moved
out of base.py verbatim. Their only dependency is gateway.session.SessionSource; base.py
imported helpers.py at module level, so helpers could not name MessageEvent. Now
TextBatchAggregator is typed by the real MessageEvent. 249 importers repointed
(`from gateway.platforms.base import` -> `.event`, preserving each import's layout);
gateway.platforms.__init__ re-exports from .event. The three revert-scheduled PLUGIN-COMPAT
pointers that named these symbols (gateway.slash_commands → MessageType, dingtalk → MessageType,
photon → ProcessingOutcome) and their COMPAT_MANIFEST rows now target gateway.platforms.event.
Docs updated: ADDING_A_PLATFORM.md, adding-platform-adapters.md (en + zh-Hans).
tools/mcp_tool_sampling.py: ElicitationHandler no longer holds a back-reference to its
MCPServerTask (mcp_tool imports sampling, so the task type cannot be named there). It only
ever read owner._pending_call_context, so it takes `call_context: Callable[[], Context | None]`
and MCPServerTask passes `lambda: self._pending_call_context`. The consent call is one
`functools.partial`, run directly or inside the captured Context.
ty on the 11 touched production files vs origin/main: 0 new diagnostics, 14 resolved.
(The one `source: SessionSource = None` diagnostic moves with the class; typing it Optional
exposes ~60 unguarded call sites — separate follow-up.)
Tests: tests/gateway + tests/plugins + tests/tools + touched files, 18,235 passed; the 31
failures reproduce identically on origin/main (macOS /private/tmp, systemd socket,
long-path fixtures, live-service tests).
Sweep of `ruff check . --select F821 --target-version py311`: 2,234 hits. 2,201 are left
alone on purpose: tui_gateway (2,169; bind_module rebinds bodies onto server.py globals,
all names verified to resolve there), the Feishu adapter (27; globals().update() SDK
binding) and the godmode script (5; dead standalone script). The other 33 were all
genuine defects. No lint config change; no TYPE_CHECKING escape hatches — every
annotation names a real, imported type; ty on the touched files: 0 new diagnostics.
- gateway/slash_commands.py: HISTORY_UNREADABLE never imported after #102117
→ NameError on the /btw error branch (same one-liner as #102952).
- gateway/platforms/whatsapp_common.py: `-> Path` return annotation with no Path
import (the body uses `_Path`). Never raised at runtime thanks to
`from __future__ import annotations`, but `typing.get_type_hints()` and ty
both fail on it.
- gateway/run.py: ActivityProvenance imported at module level
(agent.session_activity has no gateway deps); stringly annotation and the
lazy in-function import are gone.
- tools/patch_parser.py: PatchResult imported at module level; real return
annotation. The "avoid circular import" lazy import guarded a cycle that
does not exist (file_operations_common never imports patch_parser).
- gateway/platforms/helpers.py: base.py imports helpers at module level, so
MessageEvent cannot be named here; TextBatchAggregator only reads .text and
.source, so it is typed by a BatchableEvent Protocol that MessageEvent
satisfies structurally.
- tools/mcp_tool_sampling.py: mcp_tool imports this module, so MCPServerTask
cannot be named here; ElicitationHandler only reads
owner._pending_call_context, typed by an ElicitationOwner Protocol.
- plugins/platforms/sms/adapter.py: aiohttp is an optional dep ([messaging] extra) →
module-level try/except ImportError binding `aiohttp = web = None`, the pattern the
homeassistant / webhook / whatsapp_cloud adapters already use. Retires three lazy
in-function imports and the `_aiohttp_available()` wrapper; `_handle_webhook` typed
`web.Request -> web.Response`.
- plugins/platforms/teams/summary_writer.py: plain module-level `import httpx` — httpx is a
hard core dependency (pyproject `httpx[socks]==0.28.1`), so the lazy import and the
"imported on every CLI start" docstring premise were both wrong (plugin discovery never
imports this module; it is reached only via the Teams adapter / meeting pipeline).
Tests:
- tests/hermes_cli/test_config.py: a test body orphaned by the wave-1 prune
(6b81590c55) sat inside the class as dead code with self/tmp_path unbound
— header restored, so the v11→12 custom_providers migration is covered.
- tests/tools/test_mcp_tool.py: @staticmethod recursing on `self` in the
win32 branch; call portalocker directly.
- tests/test_background_review_list_shapes.py: main() still ran 3 pruned tests.
- tests/agent/test_cursor_optimizations_parity.py: bench() used names only
imported inside a sibling test.
- GatewayRunner / FeishuAdapter / Dict / Optional: missing imports.
``amount_usd == Decimal("5.450025")`` and ``pricing_version == "openai-gpt-6-astra-2026-09"``
fail on the next price or version bump without catching a bug. The whole-request tier is the
contract: above 272K prompt tokens every component — including cache writes, the field this PR
adds — is billed at its ``*_above`` rate, so derive the expected total from the entry's own rates
and assert below < above. Likewise the builtin-metadata test keeps only what
``_UNKNOWN_MODEL_BASE`` could not have supplied (vision) plus the openai-api == openai parity.
``gated_cache`` bypassed both the fresh-hit and the stale-while-revalidate branches of
``cached_provider_model_ids`` whenever the disk entry contained Astra. For an entitled
account that is every entry: live discovery rewrites the entry with Astra → next call is gated
again → a blocking /models round-trip on every picker open, for exactly the providers the user
is most likely on. The parallel prefetch's staleness check didn't know about the gate either, so
it skipped the slug and the blocking call landed in the serial picker loop the prefetch exists to
avoid.
Gate on provenance, not contents: only live discovery ever writes Astra into a same-credential
entry (the static/offline paths filter it out), so a fresh entry IS the entitlement record. The
one filter that matters stays — a stale entry served because the refresh failed drops Astra, and
the entry itself is left intact so the next successful fetch restores it.
The regression test now pins both halves: fresh entry served with Astra and zero live calls;
failed refresh past the fresh window serves the entry minus Astra.
The slug pair {"gpt-6-astra", "gpt-6-astra-900k"} was spelled out five times
(reasoning_effort, transports/codex, models.py x2, codex_models) with five hand-rolled
``.strip().lower().rsplit("/", 1)[-1]`` normalisations. ``is_astra_model`` in
agent/reasoning_effort.py (the module the other four already import) is now the only home, so a
new Astra alias is one edit.
``_finalize_codex_models(..., allow_astra=)`` was a control-coupling flag set True by the two
live callers and left False by the two static ones. The filter now lives where the untrusted
inputs are — ``_drop_undiscovered_astra`` over config.toml default + models_cache.json in
``get_codex_model_ids`` — and ``_finalize_codex_models`` is back to its one-line original.
DEFAULT_CODEX_MODELS never contains Astra, so the static catalog needs no gate.
``_openai_catalog`` appends whatever Astra ids live discovery returned instead of a name-keyed
``if "gpt-6-astra" in live_lower`` branch with a literal fallback that could never fire.
Every direct-API (api.openai.com) Astra request raised
``TypeError: Responses.create() got an unexpected keyword argument 'prompt_cache_options'``
before reaching the network: openai 2.24.0's Responses.create has no such parameter and no
**kwargs, and neither send path relocates it into extra_body. The PR's tests stopped at
build_kwargs/preflight so the SDK boundary was never crossed.
OpenAI's prompt-caching guide states ``prompt_cache_options.ttl`` accepts only ``30m`` and that
``30m`` is the default, so the field carried no information: sending nothing yields the same
cache lifetime. The sanitizer now only removes what the API rejects (none/minimal effort,
sampling/logprob knobs, the pre-5.6 ``prompt_cache_retention``) and never adds a field, which
also keeps the request body byte-stable for the cache prefix.
Also: none/minimal→low no longer needs a bespoke {"", "none", "disabled", "off"} set —
``clamp_effort`` against CODEX_ASTRA_EFFORTS already resolves to the floor (``low``); and the
auxiliary adapter derives ``is_codex_backend`` from ``classify_responses_route`` (the declared
single owner of that predicate) instead of re-implementing the host test inline.
Tests reshaped to contracts: the two proxy/subdomain cases collapse into one parametrised
"exact host only" test asserting effort and temperature pass through untouched.
Retain the two context-variant metadata additions by Michael Steuer. Keep the dedicated Astra reasoning contract already present in #103057 rather than replacing it with the GPT-5.6 vocabulary.
(cherry picked from commit add3a4fa6f31ea3f5fdad701a840d4f8530eb30e)
(cherry picked from commit 1672f12c260260ea38ed636528c02aee734d62b1)
Skip checkpoint evaluation when a turn is interrupted so cancellation rows
remain durable without urging continued execution. The existing minimal-agent
interrupt regression also avoids dereferencing an absent iteration budget.
Consolidate the warning coverage into two invariants, including real SQLite
readback and dispatcher/child scope controls. Cold-start tool availability
between construction cases to model independent worker processes. Place ratio
normalization beside the existing iteration budget instead of growing init.
Real cancelled-tool A/B against current main, draft, and fix: three cancelled
rows and zero writes on all arms; persisted checkpoint notices 0 / 1 / 0.
Repeated scripted HTTP/SQLite loop A/B preserves completion opportunity,
ordinary default-off behavior, and blocked/two-failure exhaustion behavior.
Local targeted run initially passed 15 cases with one fixture cache-isolation
failure; corrected target and inherited affected suites remain queued behind
the campaign lock. This commit is not a CI-green or merge-ready claim.
Give dispatcher-owned workers a tool-capable reporting opportunity before the
hard iteration cap, without accepting arbitrary diffs or weakening failure
counting. Add opt-in per-turn iteration checkpoints for ordinary agents.
Persist checkpoint text with the fresh tool result, never rewrite cached rows.
Salvages the opt-in ratio and per-turn reset implementation from #104683;
credits the earlier default-off signpost proposal in #92438.
Local fixture wire A/B: Kanban ready/1 failure -> done/0; deliberately stuck
workers still reach blocked/2 after two runs. Default-off control unchanged.
Targeted and affected-directory suites queued behind campaign test lock.
Co-authored-by: fangliquanflq <fangliquan@qq.com>
Co-authored-by: C. Michael Gibbs <252231331+MikeGibbsOnyx@users.noreply.github.com>
Persist paused state, timestamp, reason and no first trigger in the original
locked creation write. Forward the same boolean contract across CLI, tool,
gateway API and dashboard API, validating at the store boundary. Preserve
explicit operator force-run behavior and normal enabled creation.
The live CLI probe also caught the command shim dropping failure return codes;
forward them so invalid creation reports exit 1 rather than success.
Credit earlier atomic-creation work in #78935 and #94952 and the focused
implementation in #104578. The broader manifest staging layer is not imported.
Co-authored-by: Konstantin Khlopkov <konstantin.khlopkov93@gmail.com>
Co-authored-by: Chloé DuPont <321112755+misschloedupont@users.noreply.github.com>
Rebasing onto main surfaced three reds in tests/test_tui_gateway_server.py.
Two were a real double-check: _run_post_turn_followups already drains the
completion queue through _session_owns_notification_event, then handed the
events to _notif_handle_ready, which re-ran the belongs-elsewhere and
requires-owner gates on every event — a second compression-lineage DB
lookup per notification for events the caller had just proven ours.
_notif_handle_ready/_notif_handle_event take an explicit owned=True from
the post-turn path and skip only the ownership gates (consumed/dedup/turn
claiming still run). The poller path is unchanged (owned defaults False).
The third, test_run_prompt_submit_requeues_all_unstarted_notifications_with_real_threading,
asserted that three consecutive completions produce one in-flight turn plus
two requeued events; under this PR's contract (#104671) consecutive
completions share one turn, so the first event is now a watch_match — the
documented turn barrier — and the test keeps proving that the two
completions behind an in-flight notification turn are never lost.
A/B: revert the tui_gateway hunks and the two lineage tests fail with
ownership_checks recorded twice; restore and all 11 test_run_prompt_submit
tests pass.
Based on the earliest batching proposal by BrunoBza (#104686) and JoaoMarcos44 structured correction (#104703). Slim redo into topical siblings and shared TUI poller/post-turn routing. Reported by Xipong (#104671). Local shell-to-loopback probes demonstrate 12 to 1 turn dispatches; suites await the campaign test lock.
Co-authored-by: joaomarcos <joaomarcosdias444@gmail.com>
Capture the durable parent session before output readers start, including CLI
and non-notifying spawns. Require that parent or its compression continuation
for retained reads; exact and prefix handles alone do not authorize access.
Live Linux terminal/one-shot linger/fresh-reader A/B: base loses results;
updated owner recovers both streams and exit 7. Unbound, foreign session,
delegated child, and other profile cannot recover the receipt. No notifications
are replayed. Full tools suite is queued behind the campaign test lock.
Follow-up to contributor salvage #104805 for #104511.
Normalize incoming reasoning at the shared heading boundary and completed
extraction, and flatten auxiliary content and reasoning before accumulation.
Reuse the existing text flattener with no implicit fragment separators.
Combine the earliest related work from zsuroy (#85791), the diagnosis and
patch from 2025hcsmile2010-hue (#104711, #104848), and completed extraction
work from liuhao1024 (#104717) as a slim redo, not a verbatim cherry-pick.
Two invariant tests exercise the real SDK and local HTTP fixture across
main streaming, Relay collection, auxiliary sync/async and completed output.
The standalone matrix improves from 32/84 to 84/84, preserving answers.
Co-authored-by: suroy <suroy@qq.com>
Co-authored-by: 2025hcsmile2010-hue <2025hcsmile2010@gmail.com>
Co-authored-by: liuhao1024 <sunsky.lau@gmail.com>
`hermes_cli/banner.py::_git_run` is the shared spawn path for every banner and
passive update-check git probe (rev-parse, rev-list, remote get-url, ls-remote).
It carried the UTF-8 text contract but not `creationflags=windows_hide_flags()`,
so on Windows each probe run from a GUI-hosted backend (desktop-spawned
`hermes serve`, `tui_gateway` import kicking off `prefetch_update_check`) flashes
a console window. Every other short-lived helper in `hermes_cli/` already passes
the flag; this brings the banner path in line.
Surfaced by CI on this PR: the stray `git ... origin` spawn from the prefetch
daemon landed in `test_env_probe_run_hides_console_window`'s process-wide
`subprocess.run` capture and tripped its call-count assertion. The test now
scopes its assertion through the module's `_spawns` helper like its siblings,
so an unrelated daemon spawn cannot fail it (the production fix is what makes
that stray spawn carry the flag in the first place).
A/B: base `_git_run` -> no `creationflags` kwarg; fixed -> 0x08000000.
Ayush Nangia