teknium1 7653424356 fix(anthropic): Opus 5.5 keeps thinking on, Sonnet 5.5 turns it off with between_tools ...

Reasoning off sent thinking: {type: disabled} to both. Anthropic documents
that Opus 5.5 always thinks and 400s on disabled, and that Sonnet 5.5 400s
on disabled with a message pointing at between_tools, its lowest setting.
Every /reasoning none turn and every auxiliary call that disables thinking
(titles, compression) failed on these models.

Opus 5.5 joins Fable in keeping the omission; Sonnet 5.5 sends
between_tools. Sonnet 5 still takes the plain disable.

2026-10-04 18:29:01 -07:00

..

lsp

…

monitoring

…

pet

fix(gateway): deliver and bound-sync generated media in cache/generated

2026-10-02 01:18:03 +05:30

proxy_sources

…

secret_sources

…

transports

fix(codex): send the app-server service tier Hermes requests, only when it changes

2026-10-03 16:29:34 +05:30

vault_backends

…

verify

…

__init__.py

…

account_usage.py

…

acp_openai_bridge.py

…

activity_tracking.py

…

agent_init.py

fix: agent-start left-core migration tolerates agents without a startup-warning sink

2026-10-04 00:08:20 -07:00

agent_runtime_helpers.py

fix(compression): preserve exact coverage through composed repairs

2026-10-03 18:17:35 +05:30

AGENTS.md

…

anthropic_adapter.py

fix(anthropic): Opus 5.5 keeps thinking on, Sonnet 5.5 turns it off with between_tools

2026-10-04 18:29:01 -07:00

anthropic_credentials.py

fix(auth): find the Claude CLI outside PATH in external-process and setup-token checks

2026-10-04 17:51:24 -07:00

anthropic_endpoints.py

…

anthropic_message_convert.py

fix(anthropic): price the carrier the converter actually replays

2026-10-02 00:18:00 +05:30

anthropic_thinking_policy.py

fix(anthropic): future Claude ids keep prior thinking by default

2026-10-02 00:18:00 +05:30

anthropic_thinking_replay.py

fix(anthropic): future Claude ids keep prior thinking by default

2026-10-02 00:18:00 +05:30

api_error_summary.py

…

api_request_hooks.py

…

async_utils.py

…

aux_accounting.py

…

auxiliary_async_rebuild.py

…

auxiliary_call_outcome.py

fix(telemetry): record auxiliary model calls with their real outcome

2026-10-03 13:38:29 -07:00

auxiliary_client.py

fix(telemetry): record auxiliary model calls with their real outcome

2026-10-03 13:38:29 -07:00

auxiliary_fallback_recovery.py

…

auxiliary_health.py

…

auxiliary_hooks.py

…

auxiliary_local_runtime.py

fix(fallback): a bare llama.cpp alias resolves only to the live managed server

2026-10-02 16:44:56 -07:00

auxiliary_reasoning_floor.py

…

auxiliary_structured_output.py

…

auxiliary_unavailable.py

fix(fallback): a bare llama.cpp alias resolves only to the live managed server

2026-10-02 16:44:56 -07:00

auxiliary_wire.py

…

azure_identity_adapter.py

…

backend_identity.py

…

background_review.py

…

battery.py

…

bedrock_adapter.py

…

billing_links.py

…

billing_usage.py

…

billing_view.py

…

bounded_response.py

…

browser_provider.py

…

browser_registry.py

…

chat_completion_helpers_relay.py

…

chat_completion_helpers.py

refactor(codex): share response socket walk and simplify post-terminal drain

2026-10-03 16:31:39 +05:30

chat_completion_nonstream.py

…

chat_completion_stream_monitor.py

…

chat_completion_wait_notice.py

…

client_lifecycle.py

…

codex_headers.py

…

codex_responses_adapter.py

fix(agent): one rejected reasoning blob no longer ends replay for the session

2026-10-01 12:08:13 -07:00

codex_runtime_history_seed.py

…

codex_runtime.py

test(codex): give the no-socket drain-skip test deterministic teeth

2026-10-03 16:31:39 +05:30

coding_context.py

fix(git): neutralize repository-named filters in the session snapshot and subagent worktrees

2026-10-02 01:24:41 +05:30

command_token_source.py

…

compaction_display.py

…

compression_facade.py

refactor(compression): forward snapshot_is_current directly in facade

2026-10-03 16:54:06 +05:30

compression_marker.py

refactor(agent): share the marker count fragment and retarget the old regex comment

2026-10-02 00:50:20 +05:30

context_breakdown.py

…

context_compressor_summary.py

…

context_compressor.py

docs(compression): drop stale split-turn clause from task-snapshot comment

2026-10-04 12:02:15 +05:30

context_engine.py

…

context_file_sources.py

…

context_pin.py

…

context_references.py

refactor(context-refs): read HOME_CREDENTIAL_DIRS directly for the @file block

2026-10-04 13:53:34 +05:30

conversation_compression_archive.py

fix(compression): a row the repair drops is named, not a reason to fall back to the watermark

2026-10-03 18:17:35 +05:30

conversation_compression_manual.py

fix(compression): publish the /compress here N tail with a rotated child

2026-10-03 18:22:19 +05:30

conversation_compression_reply_anchor.py

…

conversation_compression.py

refactor(compression): hoist the commit's helper imports above the in_place split

2026-10-03 18:22:19 +05:30

conversation_loop.py

fix(agent): count restarts since the last response toward the redirect cap

2026-10-02 00:50:20 +05:30

copilot_acp_client.py

…

credential_persistence.py

…

credential_pool_admin.py

…

credential_pool_model_cooldowns.py

…

credential_pool_plugin.py

…

credential_pool.py

docs(fallback): fold the stranded #20465 comment into its branch; keep the pool docstring caller-free

2026-10-02 00:03:01 +05:30

credential_sources.py

…

credits_tracker.py

…

curator_backup.py

…

curator.py

…

deadline.py

…

delegation_context.py

fix(kanban): fence the explicit-board override to unfenced callers

2026-10-02 22:21:27 -05:00

display.py

…

empty_response_guard.py

…

error_classifier.py

fix(classifier): classify a retired free-route 404 as model_not_found

2026-10-02 22:59:00 -05:00

error_surface.py

…

errors.py

…

estop.py

…

fallback_cooldown.py

…

fast_mode.py

refactor(tui_gateway): keep the session.create tier parse out of the oversized handler module

2026-10-04 02:12:57 +05:30

file_safety.py

fix(profiles): keep .docker/.azure/.config/gh/.config/gcloud out of profile exports

2026-10-04 13:53:34 +05:30

gemini_native_adapter.py

…

gemini_schema.py

…

history_commentary.py

…

i18n_languages.py

…

i18n_layers.py

…

i18n.py

…

image_eviction_policy.py

…

image_gen_provider.py

fix(gateway): deliver and bound-sync generated media in cache/generated

2026-10-02 01:18:03 +05:30

image_gen_registry.py

…

image_routing.py

docs(image-routing): stop claiming the reverted provider-wide supports_vision probe

2026-10-04 17:54:05 -07:00

image_token_cost.py

…

inline_tool_executors.py

…

insights.py

…

interrupt_compat.py

…

interrupt_control.py

…

interrupt_scope.py

…

iteration_budget.py

…

jiter_preload.py

…

kanban_stop.py

fix(kanban): gate kanban_schedule like the other lifecycle tools

2026-10-01 09:26:00 -04:00

kanban_turn_recovery.py

…

lazy_forward.py

…

learn_prompt.py

…

learning_graph_render.py

…

learning_graph.py

…

learning_mutations.py

…

legacy_cli.py

…

lmstudio_reasoning.py

…

manual_compression_feedback.py

…

markdown_tables.py

…

memory_manager.py

…

memory_provider.py

…

message_content.py

…

message_metadata.py

fix(compression): a row the repair drops is named, not a reason to fall back to the watermark

2026-10-03 18:17:35 +05:30

message_sanitization.py

fix(agent): write normalized call_id through transport ToolCall provider_data

2026-10-02 12:29:16 -04:00

micro_compaction.py

…

moa_alternation.py

…

moa_loop.py

fix(agent): stamp turn-liveness activity from the MoA reference fan-out

2026-10-02 20:28:55 -05:00

moa_trace.py

…

model_metadata_http.py

…

model_metadata.py

fix(agent): skip /v1/models/{model} probe for unrecognised server types

2026-10-02 20:29:34 -05:00

models_dev.py

…

moonshot_schema.py

…

native_compaction.py

fix(agent): keep user image parts through typed Responses preflight

2026-10-01 16:18:23 +05:30

notification_presentation.py

…

nous_rate_guard.py

…

nous_wire.py

…

onboarding.py

…

oneshot_footprint.py

…

oneshot.py

…

opencode_affinity.py

…

outbound_webhooks.py

…

periodic_scheduler.py

…

plan_prompt.py

…

plugin_llm.py

…

plugin_stream_hooks.py

…

portal_tags.py

…

process_bootstrap.py

…

prompt_builder.py

refactor: remove Skill Sync (hermes sync) and the org skill mirror from core

2026-10-03 23:05:48 -07:00

prompt_cache_boundary.py

…

prompt_cache_scope.py

…

prompt_caching.py

…

provider_base.py

…

provider_media.py

fix(gateway): deliver and bound-sync generated media in cache/generated

2026-10-02 01:18:03 +05:30

provider_projection.py

…

provider_registry.py

…

proxy_bypass.py

…

rate_limit_credits.py

…

rate_limit_tracker.py

…

reactions.py

…

reasoning_effort.py

fix(codex): Ultra stays codex's harness mode on the app-server

2026-10-03 16:29:34 +05:30

reasoning_params.py

…

reasoning_summaries.py

…

reasoning_timeouts.py

…

redact.py

refactor(debug): reuse agent.redact lists in support redaction

2026-10-03 21:15:17 +05:30

relay_cwd.py

…

relay_llm.py

fix(telemetry): record auxiliary model calls with their real outcome

2026-10-03 13:38:29 -07:00

relay_runtime.py

fix(telemetry): record auxiliary model calls with their real outcome

2026-10-03 13:38:29 -07:00

relay_tools.py

…

repetition_guard.py

refactor(agent): drop the stream watch's redundant pre-trim

2026-10-02 00:53:19 +05:30

replay_cleanup.py

…

retry_utils.py

…

review_engine.py

…

review_idle_queue.py

…

runtime_cwd.py

…

runtime_self_protection.py

…

sdk_transform_bypass.py

…

search_policy.py

…

secret_scope.py

…

served_model.py

…

session_activity.py

fix(cron): preserve suspend credit across idle sampling

2026-10-02 13:50:23 -04:00

session_persistence.py

…

shell_hooks.py

fix(hooks): block failed hooks without directives

2026-10-02 20:18:26 -05:00

side_question.py

…

skill_bundles.py

…

skill_commands.py

refactor(skills): one disabled rule in the prompt index; hoist prefix import

2026-10-04 02:54:56 +05:30

skill_preprocessing.py

…

skill_utils.py

refactor: remove Skill Sync (hermes sync) and the org skill mirror from core

2026-10-03 23:05:48 -07:00

ssl_verify.py

…

status_output.py

fix(memory): show every memory-provider migration outcome, ask once per update

2026-10-02 04:09:57 -07:00

stream_delivery.py

…

stream_diag.py

…

stream_single_writer.py

…

subagent_lifecycle.py

…

subdirectory_hints.py

…

subscription_view.py

…

surface_switch.py

…

system_prompt.py

fix(agent): gate execution-guidance tool lines on the session's toolset

2026-10-01 11:45:41 -07:00

terminal_approval_batch.py

…

terminal_env_provider.py

…

terminal_env_registry.py

…

think_scrubber.py

…

thinking_timeout_guidance.py

…

thread_scoped_output.py

…

title_generator.py

…

tool_call_batches.py

fix(agent): execute local tool batches through per-tool pipeline

2026-10-01 23:50:31 -07:00

tool_dispatch_helpers.py

refactor: move Home Assistant out of core into the homeassistant catalog plugin

2026-10-04 00:08:20 -07:00

tool_executor.py

Revert "docs(agent): the authorization-gate bound follows the turn's approval window"

2026-10-01 13:23:12 -07:00

tool_guardrails.py

…

tool_result_classification.py

…

trace_upload.py

…

trajectory.py

…

transcript_repair.py

fix(state): keep stored token_count when the agent flush row carries token_count=None

2026-10-02 01:04:12 +05:30

transcription_provider.py

…

transcription_registry.py

…

tts_provider.py

…

tts_registry.py

…

turn_api_call.py

…

turn_api_error.py

…

turn_api_request.py

…

turn_author.py

…

turn_context_compaction.py

…

turn_context.py

fix(agent): name a final that reuses an already-delivered response

2026-10-04 18:04:17 -07:00

turn_empty_response.py

fix(agent): name a final that reuses an already-delivered response

2026-10-04 18:04:17 -07:00

turn_explainers.py

…

turn_facade_lease.py

refactor(state): hoist pidns imports and trim duplicated call-site prose

2026-10-03 17:44:58 +05:30

turn_facade.py

…

turn_failure_copy.py

…

turn_final_response.py

fix(agent): don't promote Anthropic summarized thinking to the final answer

2026-10-04 17:47:39 -07:00

turn_finalizer.py

fix(agent): name a final that reuses an already-delivered response

2026-10-04 18:04:17 -07:00

turn_iteration_prep.py

docs(agent): describe restart_count as restarts since the last response

2026-10-02 00:50:20 +05:30

turn_liveness.py

fix(agent): stamp the activity clock at turn end and dump thread stacks on watchdog fire

2026-10-03 16:18:04 -05:00

turn_loop_errors.py

…

turn_overflow.py

…

turn_preflight_gate.py

…

turn_preflight.py

…

turn_recovery_autorecover.py

…

turn_recovery.py

fix(agent): reasoning replay verdict resets when the route changes

2026-10-01 12:08:13 -07:00

turn_request_assembly.py

fix(anthropic): simplify recovery and make accounting explicit

2026-10-02 00:18:00 +05:30

turn_response_check.py

…

turn_response_intake.py

…

turn_retry_state.py

…

turn_stop_gates.py

…

turn_summary.py

…

turn_tool_round.py

fix(agent): normalize the final staged tool batch by effective id

2026-10-02 12:29:16 -04:00

turn_tool_validation.py

fix(agent): normalize provider-minted parallel tool ids

2026-10-02 12:29:16 -04:00

turn_truncation.py

fix(agent): cut a runaway repetition loop while it streams, callback or not

2026-10-02 00:53:19 +05:30

turn_usage.py

…

usage_anchor.py

…

usage_pricing.py

refactor(pricing): generate the gpt-6.1-sol -pro/-900k aliases from the tier loop

2026-10-01 16:40:58 +05:30

vault_login_classifier.py

…

vault_store.py

…

verification_evidence.py

…

verification_stop.py

…

verify_hooks.py

…

vertex_adapter.py

…

video_gen_provider.py

fix(gateway): deliver and bound-sync generated media in cache/generated

2026-10-02 01:18:03 +05:30

video_gen_registry.py

…

vision_message_prep.py

…

web_search_provider.py

…

web_search_registry.py

…