Keep MoA reference display events off the machine-readable -Q stdout
surface (platform=cli with tool_progress_mode=off) while preserving them
everywhere else. Extracts the relay into module-level helpers so the
policy is testable.
Salvaged from #67334.
Follow-up to the #67690 salvage (@m4r13y). The PR's tools/env_probe.py
hunk was written against the old capture_output=True _run(); #67964/#67999
rewrote _run to temp-file capture on July 20, so that hunk no longer
applied — but the rewritten _run still lacked creationflags and kept
flashing one console per probe (~5 per kanban worker start) from
windowless parents. Re-implement the one-line fix against the current
shape: creationflags=windows_hide_flags() on the temp-file subprocess.run,
preserving the #67964 grandchild-can't-wedge-the-pipe contract.
Also add the tests the PR didn't ship, in
tests/test_windows_subprocess_no_window_flags.py:
- env_probe._run passes CREATE_NO_WINDOW and keeps temp-file (non-PIPE)
stdout/stderr + DEVNULL stdin
- lazy_deps uv install / pip --version probe / pip install fallback /
ensurepip bootstrap all pass CREATE_NO_WINDOW
- suppress_platform_ver_console: POSIX no-op (platform._syscmd_ver
untouched, win32_ver() still returns), and simulated-Windows stubbing
(echo stub installed, idempotent, never raises)
The per-advisor enabled toggle adds enabled=True to normalized slots;
the JSON-string-parse and per-slot max_tokens tests from sibling
clusters asserted exact dicts. Compare against the enabled-augmented
expectation instead.
The per-reference-model enabled toggle (#59753 salvage) intentionally adds
'enabled' to normalized slot dicts. The two endpoint tests asserted the
exact key set {provider, model} — convert them to subset + round-trip
contracts so optional slot keys (enabled, reasoning_effort, max_tokens)
don't break them again.
Follow-up for salvaged PR #59753 rebased over the per-slot
reasoning_effort feature: _clean_slot now round-trips reasoning_effort
AND enabled together; add a normalize→normalize regression test, update
the validate/normalize agreement contract for the canonical enabled
default, restore the desktop per-slot toggle test on the current
autosave editor, and map oppenheimor's contributor email.
Adds per-reference progress events and a phase-transition marker to the
MoA display pipeline so TUI / CLI / desktop surfaces can render a status
bar like `MOA: 2/3 refs done` and surface which phase (reference vs
aggregator) is currently active.
- `moa.progress` — fired once per reference completion with
`refs_done`, `refs_total`, and the source label
- `moa.phase` — fired on phase transitions (currently the single
`phase="aggregator"` transition once the fan-out
finishes)
Plumbed through the existing `reference_callback` →
`tool_progress_callback` → gateway path; no new UI surface. The legacy
`moa.reference` / `moa.aggregating` events are unchanged for backwards
compatibility.
AI-assisted fix by https://github.com/SquabbyZ/peaks-loop
Regression tests for the #47971 salvage: the LSP language-server spawn
must pass windows_hide_flags() creationflags while keeping PIPE stdio
and start_new_session, and the npm/go LSP auto-installer subprocess.run
calls must carry the same hide flags with DEVNULL stdin and
capture_output intact.
The Windows /restart watcher's outer Popen spawns the watcher with
windows_detach_popen_kwargs() (which carries CREATE_BREAKAWAY_FROM_JOB),
but a restrictive parent job object can reject that bit with OSError and
the current call has no retry. Preserve the current watcher
implementation and add a focused breakaway-denied fallback.
Preserved from current main: watcher_python / pythonw.exe selection, the
str(restart_after_s) deadline, the scrubbed watcher_env, the intentional
no-breakaway inline respawn, and the entire POSIX setsid/bash path.
- primary keeps **windows_detach_popen_kwargs()
- on OSError, retry the same argv/env with
creationflags=windows_detach_flags_without_breakaway()
- on dual failure, log a definitive, path-safe warning (interpreter
basename + numeric winerror/errno only) and return without crashing
Replace the superseded breakaway-first inline design and its AST tests
with focused behavioral coverage that drives the real coroutine with a
mocked subprocess.Popen (retry, argv/env/DEVNULL preservation, POSIX
single-session kwarg, no-breakaway inline respawn, secret-safe logging).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds moa.privacy_filter ('' | display | full, default off — issue #59959):
- display: redact user-visible surfaces only (reference blocks emitted to
the UI + saved MoA trace records, including per-advisor full input/output
and the aggregator-input copy); the aggregator sees raw advisor text so
synthesis quality is unaffected.
- full: additionally redact the advisor text injected into the aggregator
prompt, on both the persistent facade path and the one-shot /moa
synthesis path (the issue's literal ask). Legacy boolean true maps here.
Secret/credential shapes (API-key prefixes, JWTs, private keys, DB
connection strings) are delegated to the central redactor
(agent.redact.redact_sensitive_text, force=True + code_file=True); the MoA
filter adds only email and clearly delimited phone-number patterns. No
bare 10-digit matching: line numbers, timestamps, epoch values, git SHAs,
IPs, versions, and source-code assignments in code-review-shaped advisory
text pass through byte-identical. The reference cache always holds raw
text — redaction happens at each consuming surface, so a mid-session mode
change never leaks or double-redacts.
Reworked from PR #60463: replaced its hand-rolled pattern list (which
matched bare digit runs and re-implemented key shapes) with central-
redactor reuse + safe patterns, and split the single boolean into
display/full modes. Credited for the feature framing.
Co-authored-by: webtecnica <75556242+webtecnica@users.noreply.github.com>
Extends the fanout enum with 'every_n:<N>' (N >= 2): advisors run on the
first iteration of each user turn and every Nth tool iteration after it;
off-cadence iterations REUSE the cached guidance from the last on-cadence
run via the same cache mechanism the user_turn fanout uses, so the
aggregator still gets advice on every step. The cadence counter is scoped
per user turn (resets on a new user message) and only advances when the
advisory state actually changes, so streaming retries never consume a
cadence slot. Mapping form {mode: every_n, n: N} normalizes to the
canonical string. Unknown/degenerate values fall back to per_iteration.
Addresses issue #63393 (advisor fan-out multiplies turn latency/cost by
the tool-iteration count). Redesigned from PR #63448: the submitted shape
skipped references entirely on off-cadence iterations (aggregator ran
advice-less); this version keeps the last advice in play, credited for
the idea and cadence framing.
Config-gated, default-off (default fanout remains per_iteration).
Co-authored-by: webtecnica <75556242+webtecnica@users.noreply.github.com>
Adds test_moa_gemini_aggregator_sanitize_uses_real_model: drives a full MoA
tool-call turn (virtual-provider mode) with a Gemini aggregator and asserts
the strict-API sanitize pass is invoked with the resolved aggregator model
(gemini-3-pro-preview), never the virtual preset name once a slot is
resolved — the exact path that stripped extra_content/thought_signature and
made Gemini aggregators 400 (#65092).
Writing the test surfaced a gap in the salvaged #66212 fix: in virtual-
provider MoA mode (provider=moa, no moa_config threaded through
run_conversation) the conversation-loop branch never fired because it only
consulted moa_config. Extend it to fall back to the facade's
last_aggregator_slot — the same source the handle_max_iterations fix uses —
so both MoA entry modes resolve the real aggregator model.
Also adds the contributors/emails mapping for the #15676 credit base.
Follow-ups to the salvaged core of #60293:
- Gate the x-initiator header on _normalize_aux_provider() instead of a
literal 'copilot' string compare, so slot configs spelled github /
github-copilot / github-models / copilot-acp / mixed case all get the
user-turn attribution.
- Thread extra_headers through _retry_same_provider_sync/_async so the
credential-refresh and pool-rotation retry rebuilds don't silently drop
the header (the rebuilt kwargs previously started from scratch).
- Add a transport-boundary test asserting the header reaches the SDK
client's create() kwargs (no call_llm mocking), an alias-spelling
matrix test, and a retry-rebuild preservation test.
Follow-up to the salvaged core of #53802: a naive MoAClient(preset) rebuild
restores a working facade but silently drops the reference_callback relay
wired in agent_init, so moa.reference / moa.aggregating display events stop
reaching every frontend for the rest of the session.
Introduce agent.moa_loop.build_moa_facade(agent, preset) as the single
construction point for the MoA facade and use it at:
- initial client construction (agent_init.py)
- turn-start fallback restore (restore_primary_runtime)
- transient transport recovery (try_recover_primary_transport — previously
fell through to _create_openai_client with MoA's empty client_kwargs and
died with 'api_key client option must be set')
- mid-session model switches (switch_model)
The relay reads agent.tool_progress_callback at emit time, so callbacks
attached after construction are picked up automatically.
Adds test_moa_restored_facade_still_emits_reference_events covering event
delivery through a restored facade.
When Hermes fails over from a non-Gemini provider (xAI, Anthropic, etc.) to
Gemini mid-conversation, the existing assistant tool_calls in history carry
no Gemini ``extra_content.google.thought_signature`` (the originating provider
never emits one). The native adapter's ``_translate_tool_call_to_gemini``
omitted ``thoughtSignature`` entirely in that case, so Gemini 3 thinking
models rejected every replayed turn with::
HTTP 400 INVALID_ARGUMENT
Function call is missing a thought_signature in functionCall parts.
Additional data, function call default_api:<tool_name>, position N.
The Cloud Code Assist sibling adapter already handles this exact case by
emitting a sentinel ``"skip_thought_signature_validator"`` (see
``agent/gemini_cloudcode_adapter.py:106``, originally added in #11270 and
documented as matching ``opencode-gemini-auth``'s approach). This change
mirrors that fallback in the native adapter so the two paths behave
identically when replaying cross-provider history.
Verified live against ``generativelanguage.googleapis.com/v1beta`` with
``gemini-3-pro-preview``: synthetic 2-turn conversation with no real
``thoughtSignature`` returns 400 without the sentinel and 200 with it.
Test added: ``test_build_native_request_emits_sentinel_for_cross_provider_tool_call``.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Follow-up fixes on top of the salvaged #22566 mechanism:
- N-collector now counts only REAL actionable user turns via
_is_actionable_user_turn + _is_synthetic_compression_user_turn —
the same filter pair _find_last_user_message_idx uses post-#69291.
The contributor's bare role=='user' + _is_context_summary_content
check let blank platform echoes and continuation/todo rows consume
N slots, silently degrading the guarantee.
- Default flipped 3 -> 1 (behavior-preserving): a default of 3 was
measured to change the tail cut on transcripts whose budget covers
only the last turn. min_tail_user_messages=1 delegates to the
existing single-user anchor; N>1 is opt-in, and the call site is
gated so the default path is byte-identical to main.
- Hardened config parse in agent_init (bool rejected, fractional
floats rejected, floor 1) matching the max_attempts parser shape.
- Wired the recurring external-PR config gaps: hermes_cli/config.py
DEFAULT_CONFIG + cli-config.yaml.example (PR only had cli.py).
- Regression tests: blank echoes / synthetic rows don't count toward
N; tool-call/result pairs never split by the N-boundary (no-orphan
both directions); N-guarantee wins over tail_token_budget and the
_MAX_TAIL_MESSAGE_FLOOR (floor is a minimum, not a cap); default
parity pin; DEFAULT_CONFIG pin.
Add _ensure_last_n_user_messages_in_tail to guarantee the last N user
messages survive compression in the uncompressed tail, with surrounding
assistant/tool context preserved.
- Add min_tail_user_messages parameter (default 3) to ContextCompressor
- New _ensure_last_n_user_messages_in_tail method generalizes single-user protection
- Skip context-summary handoff banners when counting user messages
- User messages are clean boundaries — skip _align_boundary_backward
- Wire through cli.py, agent_init.py, and gateway cache busting keys
Config:
compression:
min_tail_user_messages: 3
Co-Authored-By: Claude <noreply@anthropic.com>
_collect_ghosted_skill_names() covers both ghost-skill shapes in the
compressed middle window: rows already demoted to a [SKILL_PRUNED: ...]
marker AND raw skill_view bodies (> _SKILL_VIEW_PRUNE_MIN_CHARS) that
survived Phase-1 inside an earlier protected tail and then aged into the
compression window — the summarizer paraphrases those instructions away
too. Shared threshold constant between the emit site and the scan.
Pinned by a live-probe-shaped test (real compress(), mocked aux LLM).
21 tests pinning the salvaged #44166 behavior:
- marker emit + extractor round trip (patterns adapted from PR #32375
by @LeonSGP43, with credit)
- no-duplicate re-injection when the canonical marker survived (the
original PR's presence-check defect)
- Phase-1 protection for just-loaded / user-referenced skills, and the
Pass-4 pressure override that keeps #61932 fixed
- deterministic marker survival through a REAL compress() with a mocked
aux LLM: drop → re-injected, keep → not duplicated, static-fallback
path, iterative re-compression via rehydrated handoff
- markers never classify as handoff content (classify_summary_content /
_strip_context_summary_handoff_message untouched)
- SKILLS_GUIDANCE Skill Safety Rule renders with real newlines
Community verification of #56688 (zmack12344321) found two follow-up gaps
that kept Vertex invisible in the /model menu even after registry
registration:
1. hermes_cli/model_switch.py: list_authenticated_providers() had a
credential gate hard-coded to API keys (with an aws_sdk special case
only) — add a vertex branch using has_vertex_credentials(), mirroring
the aws_sdk shape.
2. hermes_cli/models.py: Vertex's OpenAI-compatible endpoint has no
/models listing route, so without a curated _PROVIDER_MODELS entry the
picker only ever showed the current model — add a Gemini curated list.
Follow-up to #56688.
The Vertex AI provider (added same-day, commit c73e74386) was never added to
either of the two provider registries that agent/auxiliary_client.py and the
MoA slot-resolution chain depend on, breaking Vertex outside the main
conversation loop:
1. hermes_cli/auth.py::PROVIDER_REGISTRY had no "vertex" entry. The
plugin-auto-extend loop that normally fills gaps explicitly skips
non-api_key auth types (`if _pp.auth_type != "api_key": continue`), and
Vertex was never hand-declared like "bedrock" is. Because
resolve_provider_client() in agent/auxiliary_client.py gates everything
on `pconfig = PROVIDER_REGISTRY.get(provider)` and returns (None, None)
immediately when pconfig is None, its `elif pconfig.auth_type == "vertex"`
branch was permanently dead code — every auxiliary Vertex call (vision,
title generation, reflection, context compression, MoA reference/
aggregator slots) failed outright, not just a MoA-specific edge case.
2. hermes_cli/providers.py::HERMES_OVERLAYS also had no "vertex" entry, so
hermes_cli.providers.get_provider("vertex") returned None. This backs
_preserve_provider_with_base_url() in agent/auxiliary_client.py, which a
MoA slot's resolved (base_url, api_key) pair needs to keep its "vertex"
identity instead of silently collapsing to "custom" — losing the
identity _refresh_provider_credentials() needs to re-mint an expired
OAuth2 token (~1h lifetime) on a 401, and permanently breaking every
subsequent call in that MoA preset for the rest of the session.
Fix mirrors the existing "bedrock"/aws_sdk entries in both registries
exactly, plus adds a "vertex" branch to _refresh_provider_credentials() (it
had branches for openai-codex/nous/anthropic/xai-oauth but not vertex,
so a 401 fell through to `return False` without evicting the stale cached
client).
- hermes_cli/auth.py: hand-declared vertex ProviderConfig(auth_type="vertex")
in PROVIDER_REGISTRY, matching bedrock's shape.
- hermes_cli/providers.py: vertex HermesOverlay(auth_type="vertex") in
HERMES_OVERLAYS + "Google Vertex AI" label override.
- agent/auxiliary_client.py: vertex branch in _refresh_provider_credentials
that re-mints the token via get_vertex_config() and evicts the stale
cached client.
- 8 new regression tests across tests/hermes_cli/test_vertex_provider.py and
tests/agent/test_auxiliary_client.py: registry membership, end-to-end
resolve_provider_client("vertex", ...) building a working client (proving
the previously-dead branch is now reachable), and the 401-refresh/cache-
eviction path.
MoaConfigPayload does not declare save_traces or trace_dir, so
set_moa_models() overwrites cfg["moa"] with a dict that lacks these
hand-edited keys. Use dict.update() to merge instead of replace.
Fixes#58819
MoA (Mixture of Agents) is a legitimate internal provider used by
Diagnosis presets and multi-model aggregation. When a MoA preset sets
model.provider to 'moa', hermes doctor incorrectly reports it as
'unrecognised' and suggests changing it, which would break the MoA setup.
Add 'moa' to the known_providers set alongside 'openrouter', 'custom',
and 'auto' so doctor recognises it as valid.
Fixes#58759
The existing tests cover _normalize_moa_model() in isolation and a local
precedence expression, but not the __init__ wiring itself. Add two
init-level regression tests: constructing HermesCLI(model='moa:strategy')
strips the prefix to model='strategy' and forces requested_provider='moa',
and the moa: prefix wins over an explicit --provider. Both fail if the
override is dropped from the requested_provider resolution.
Refs #56828
hermes chat -Q -m moa:strategy failed with 'model moa:strategy is not
supported' (HTTP 401/400): the raw model string was passed straight to
the real provider. The MoA virtual provider only got wired up through the
interactive /moa command and the model picker, never through the -Q
one-shot startup path.
resolve_runtime_provider already handles requested_provider == 'moa', and
agent_init builds the MoAClient off provider == 'moa' (surface-agnostic).
The only gap was mapping the moa:<preset> model string to that provider.
Add _normalize_moa_model() and apply it in HermesCLI.__init__ before
provider resolution: a moa:<preset> model sets requested_provider='moa'
and model=<preset>, so the existing MoA path runs in non-interactive mode
too. The moa: prefix wins over an explicit --provider (previously
--provider deepseek -m moa:strategy silently dropped MoA).
Fixes#56828
Follow-ups on top of the cherry-picked #27748 mechanism:
- move the cap constant to module level with full rationale comment
(class attribute aliases it so subclasses/tests can override)
- bound the iterative-update path too: the PREVIOUS SUMMARY block is
passed through _bound_summary_input so a pathological rehydrated
handoff cannot blow up the prompt (previous summary + new turns each
capped)
- extra regression tests: byte-identical small-input passthrough
(identity), direct bound+marker unit check, bound-after-per-message-
truncation shape (hundreds of under-_CONTENT_MAX turns), iterative
path bounded, marker vs classify_summary_content non-collision
- contributor email mapping for @robgfl45
Follow-ups on top of the cherry-picked #62644 mechanism, porting it to
current main and closing the salvage-review requirements:
- proactive_prune_min_reclaim_tokens (default 4096): a prune only COMMITS
when it reclaims a meaningful token batch, measured on the pruned output.
A committed prune rewrites already-sent history and invalidates the
provider prompt-cache prefix; this hysteresis gate keeps those breaks
episodic/amortized (like a compression boundary) instead of firing every
tool iteration. 0 disables the gate. (Design point credited to the
#62389 review cycle's prune_minimum_tokens.)
- Standard no-op caller contract: every skip path returns the INPUT list
object; the loop commits only on 'result is not messages' + non-zero count.
- Loop call is getattr+callable guarded (plugin engines predating the hook,
SimpleNamespace test doubles) and exception-swallowed at debug level.
- Config parse follows the compression.max_attempts hardened semantics:
booleans rejected, fractional floats rejected, integral floats/numeric
strings accepted; negative trigger = disabled.
- cli-config.yaml.example documented (all three keys) and gateway
_CACHE_BUSTING_CONFIG_KEYS extended so hot-reload rebuilds the agent.
- Tests: min-reclaim gate both directions, input-object no-op contract,
no-orphan tool_call_id pairing in BOTH directions (#69830 pin rule),
default-off zero-behavior-change pin, config parse seam, and behavioral
loop-wiring tests (consulted/commit/no-op/absent-method/raising).
The phase-1 tool-result prune only runs inside compress(), which fires
near 50% of the context window, so it never triggers on large-window
models; old tool outputs then ride in history and are re-sent every turn.
Add prune_tool_results_only(): the same no-LLM prune on a separate, low
proactive_prune_tokens trigger, run as an elif to the compression branch.
Opt-in (default 0), protects the recent tail by message count.
Add the method to the ContextEngine base as a no-op default so pluggable
engines inherit it safely (the post-tool-call path never AttributeErrors on
a non-built-in engine); the built-in compressor supplies the real prune.
Register both keys under the top-level compression config with defaults and
document them.
Interrupted turns can seed a speculative display token count before the provider receives the request. Restore that display-only seed when interruption wins the race, while preserving completed post-compaction state and treating a successful provider response independently of optional usage metadata.
Constraint: #54776 remains reproducible on current main, while review #4702305384 identifies anti-thrashing rollback as stale and usage receipt as an unreliable response-completion signal.
Rejected: Restore anti-thrashing counters from a preflight snapshot | current main derives their verdict from real provider usage after a completed compaction boundary.
Confidence: high
Scope-risk: narrow
Directive: Keep interrupted preflight rollback display-only, and never infer provider completion from the presence of usage metadata.
Tested: ./.venv/bin/python -m pytest -q tests/run_agent/test_413_compression.py (29 passed); turn-finalizer/conversation-loop tests (31 passed); context-compressor targeted tests (12 passed); infinite-compaction targeted tests (3 passed); ruff; git diff --check.
Not-tested: End-to-end interactive interrupt through CLI or gateway transport.
Gateway half of the #49874 salvage: pass compression_deferred through
both _run_agent_inner result dicts and guard the compression-exhausted
auto-reset block with it — a lock-contended defer keeps the session
intact (the concurrent compressor is actively shrinking it) instead of
wiping it via reset_session.
Regression tests:
- tests/run_agent/test_compression_lock_defer.py — provider-mock 413 and
400-overflow turns whose compression pass lost the lock end as
compression_deferred (failed=False, no compression_exhausted); flag
unset keeps the terminal exhaustion path byte-identical; type-pin
tests vs MagicMock agents and junk flag values; cap=1 e2e proving the
refunded pre-API defer leaves the budget for the provider-proven
413 retry.
- tests/agent/test_preflight_lock_defer.py — a lock-skipped preflight
pass stops the loop WITHOUT arming preflight_compression_blocked;
plain no-op still arms it; MagicMock junk does not defer.
- tests/gateway/test_compression_deferred_soft_result.py — AST pin that
the deferred branch guards the auto-reset chain and performs no
session mutation (mirrors test_35809_auto_reset_clean_context.py).
Hardening on top of the salvaged dead-PID lease reclamation from PR #65775
(@the3asic):
- Probe via psutil.pid_exists (hard dependency; CONTRIBUTING.md critical
rule #1) with the contributor's os.kill(pid, 0) POSIX probe retained
only as a scaffold-phase fallback when psutil is missing.
- Same-process holders (pid == os.getpid()) are never probed and never
self-reclaimed — another thread's live lease is owned by the lease
refresher/release path.
- Any probe doubt (exceptions, permission errors) conservatively keeps
the lease until normal TTL expiry; Windows stays TTL-only.
- Tests: psutil-first dead-pid reclaim (probe call pinned), os.kill
fallback path, probe-doubt keeps lease, same-pid no self-reclaim,
legacy holder + Windows paths assert NO probe via either API.
compress_trajectory (and _async) replaced the compressible middle region with
a [CONTEXT SUMMARY] turn without checking that the region is actually larger
than the summary. When a large protected system prompt dominates the budget,
the compressible middle can be tiny; replacing e.g. a 2-token middle with a
~60-token summary GROWS the trajectory (tokens_saved negative), marks it
was_compressed, and still spends a summarization call — the opposite of the
intent, on exactly the hard over-budget cases.
Add a net-savings guard mirroring the code's own comment (net_savings =
region_tokens - summary_target_tokens): if the safely-compressible region is
no larger than summary_target_tokens, return the trajectory unchanged. Applied
to both the sync and async paths. Add sync+async regression tests.
MoA reference_max_tokens is preset-level — one cap for all reference
models. When mixing a verbose model with a terse one, a single cap is
either too tight for the terse model or too loose for the verbose one.
Now each reference slot can optionally carry its own max_tokens:
reference_models:
- provider: openrouter
model: deepseek/deepseek-v4-pro
max_tokens: *** # per-slot cap, overrides preset-level
- provider: openai-codex
model: gpt-5.5
# no max_tokens → falls back to preset-level reference_max_tokens
_clean_slot (moa_config.py) preserves an optional max_tokens field on
the slot dict, coerced via _coerce_int_or_none. _run_reference
(moa_loop.py) reads slot-level max_tokens first, falling back to the
preset-level cap passed by the caller. Slots without the field are
unaffected — backward compatible.
Type hints on slot-handling functions updated from dict[str, str] to
dict[str, Any] to reflect the now-heterogeneous slot shape.
Requested in review: builder-level assertions that the gemini-native
branch forwards max_tokens (provider names and the native
generativelanguage.googleapis.com base_url, max_tokens=600), plus a
control showing gemini models on OpenAI-compatible endpoints — including
Gemini's own /openai compatibility endpoint — keep the existing omission
behavior (#34530).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Per review feedback from teknium1: reference_max_tokens is an advisors-only
contract. The aggregator is the acting model and must not be capped by the
reference budget. Changed _is_moa from startswith('moa_') to exact match on
'moa_reference'. Added regression test proving aggregator does NOT receive
max_tokens.
Copilot review pointed out that hardcoding kwargs['max_tokens'] would
400 on models requiring max_completion_tokens (GPT-5 family, Copilot).
The existing auxiliary_max_tokens_param() helper already selects the
correct parameter name per model — use it instead of hardcoding.
Test updated to parametrize expected_key so the Copilot gpt-5.5 case
correctly asserts max_completion_tokens instead of max_tokens.
Addresses Copilot review comments on both files.
PR #56756 added reference_max_tokens to cap MoA advisor output and cut
turn latency. The value is correctly threaded through five layers of MoA
code (moa_config → conversation_loop → aggregate_moa_context →
_run_references_parallel → _run_reference → call_llm(task='moa_reference',
max_tokens=800, ...)).
However, _build_call_kwargs() in auxiliary_client.py silently drops
max_tokens for all OpenAI-compatible providers (PR #34845, which fixed
endpoints and NVIDIA NIM keep it. This means reference_max_tokens never
reached the API for the vast majority of providers.
The bug affects every OpenAI-compatible MoA reference/aggregator slot:
Z.AI (coding plan), OpenRouter, OpenAI, GitHub Copilot, and local
providers. Only Anthropic-compat endpoints (MiniMax, /anthropic URLs)
worked — by coincidence, not MoA-aware design.
Fix: thread the 'task' parameter through all six _build_call_kwargs()
call sites. When task starts with 'moa_', max_tokens is always included
in the request kwargs regardless of provider. Non-MoA auxiliary tasks
(compression, titles, vision, etc.) keep PR #34845 behavior unchanged.
Verified end-to-end:
- Z.AI GLM-5.2 with max_tokens=50 → returned exactly 50 tokens
- Z.AI GLM-5.2 with max_tokens=20 → returned exactly 20 tokens
- Z.AI GLM-5.2 uncapped → returned 315 tokens
- 7 new regression tests covering 4 providers, Anthropic wire, non-MoA
tasks, and prefix-matching boundary
- 288 auxiliary_client tests pass (was 281, +7 new), 84 MoA tests pass
- Zero regressions
aggregate_moa_context's single max_tokens parameter was applied to
both the reference fan-out (_run_references_parallel) and the
aggregator's own synthesis call_llm. #53580 explicitly removed a
hardcoded cap from the aggregator call because it truncated long
aggregator syntheses; #56756 (reference_max_tokens, added to speed up
the advisor fan-out) reintroduced the same shared cap by passing it to
both calls, silently regressing #53580's fix.
Rename the parameter to reference_max_tokens (matching the caller's
own moa_config key) and stop forwarding it to the aggregator's
call_llm invocation, which now always runs uncapped as intended.
An assistant message with an empty or whitespace-only text content block —
produced by context compression or certain tool-call flows — is rejected by
the Anthropic Messages API with HTTP 400 "text content blocks must contain
non-whitespace text". Because the blank block is stored in session history and
replayed verbatim every turn, the session is permanently wedged behind the same
400.
The Bedrock adapter already guards this via _safe_text() (#9486); the native
Anthropic path never got the same treatment. _sanitize_replay_block() rebuilt
text blocks with the raw stored text, and the _convert_assistant_message()
guard only caught a fully empty block list, not a list still containing a
whitespace-only text block.
Add a _safe_text() helper mirroring the Bedrock one and apply it at both points:
the ordered-blocks replay path and a final in-place walk of the converted
content list. Both are self-healing — sessions that stored blank blocks recover
on the next API call. Only text blocks are coerced; thinking/tool_use/image
blocks are untouched.
Fixes#69512
