Teknium e3d2e57c9e fix(skills): background review can improve every skill, not just curator-managed ones ...

The self-improvement review fork refused every edit/patch/write_file/remove_file on a skill
without `created_by: "agent"`, so lessons about hand-written, URL-installed, foreground-created,
bundled, hub, external or pinned skills were dropped with only a WARNING line (77% of refused
review writes on the #134289 reporter's install). The prompts then steered the fork to a
different umbrella or a new near-duplicate skill.

Ownership now gates only delete/archive: the review may improve any skill it learns from (still
read-before-write, ledgered, and staged when skills.write_approval is on), while autonomous
archival stays limited to curator-managed, unpinned, non-bundled/hub/external skills. The review
prompts drop the protected-skill lists and say never to delete.

Fixes #134289

2026-10-07 02:27:40 -07:00

..

lsp

…

monitoring

fix(monitoring): report cron store health across every served profile

2026-10-07 10:20:18 +05:30

pet

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

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

proxy_sources

…

secret_sources

fix(secret-sources): revoke a removed source's value from os.environ

2026-09-29 22:57:48 -04:00

transports

refactor(providers): one copy each of TokenHub's effort rule and the generic nested reasoning fallback

2026-10-07 11:27:35 +05:30

vault_backends

refactor(approval): one "no user can answer" helper for every CLI-callback prompt

2026-10-05 20:08:11 +05:30

verify

fix(processes): signal a process group only when the child leads it

2026-10-05 16:42:57 -05:00

__init__.py

…

account_usage_cache.py

feat(usage): report credential-pool usage per account

2026-10-05 14:02:38 -05:00

account_usage.py

feat(usage): report credential-pool usage per account

2026-10-05 14:02:38 -05:00

acp_openai_bridge.py

…

activity_tracking.py

…

agent_init_fallback.py

fix(copilot): recompute api_mode on fallback and stop stale-mode veto (#46527)

2026-10-06 23:54:51 -05:00

agent_init.py

fix(copilot): recompute api_mode on fallback and stop stale-mode veto (#46527)

2026-10-06 23:54:51 -05:00

agent_runtime_helpers_dead_connections.py

refactor(agent): move dead-connection sweep into a runtime-helpers sibling

2026-10-07 00:07:02 -05:00

agent_runtime_helpers.py

refactor(agent): extract the fallback route swap and primary restore into route_binding

2026-10-07 01:35:11 -07:00

AGENTS.md

…

anthropic_adapter.py

fix(lint): stdin guard catches calls whose arguments start on the next line (review)

2026-10-07 01:03:03 +05:30

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

feat(fast): /fast ultrafast for GPT-6 Astra, priced at the served tier

2026-09-29 23:54:34 +05:30

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_registry.py

feat(auth): brokered PKCE, expiry-driven rotation and aux on plugin OAuth rows

2026-10-05 23:33:41 -07:00

auxiliary_client.py

refactor(aux): move rotate_runtime_main_api_key into auxiliary_key_rotation to keep auxiliary_client.py under its size cap

2026-10-07 13:20:42 +05:30

auxiliary_codex_response.py

refactor(auxiliary): move the Codex aux response normalizer into a sibling

2026-10-05 19:35:01 +05:30

auxiliary_fallback_recovery.py

…

auxiliary_health.py

…

auxiliary_hooks.py

…

auxiliary_key_rotation.py

fix(aux): stop republishing the legacy mirrors from a key rotation

2026-10-07 13:20:42 +05:30

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_plugin_oauth.py

feat(auth): brokered PKCE, expiry-driven rotation and aux on plugin OAuth rows

2026-10-05 23:33:41 -07:00

auxiliary_reasoning_floor.py

…

auxiliary_stream_watchdog.py

fix(aux): the first-token window honours providers.<id>.stale_timeout_seconds

2026-10-05 16:00:18 -07:00

auxiliary_structured_output.py

…

auxiliary_task_config.py

refactor: move aux-slot growth into topical siblings (code-health caps)

2026-10-05 17:24:40 -07:00

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

fix(skills): background review can improve every skill, not just curator-managed ones

2026-10-07 02:27:40 -07:00

battery.py

…

bedrock_adapter.py

fix(bedrock): route GPT-6 family and prefixed OpenAI inference profiles, size their 1M window (#133573)

2026-10-06 08:13:51 -04:00

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(agent): extract the fallback route swap and primary restore into route_binding

2026-10-07 01:35:11 -07:00

chat_completion_nonstream.py

…

chat_completion_stream_monitor.py

…

chat_completion_wait_notice.py

…

client_lifecycle.py

refactor(compression): import the codex cooldown lookup at module level

2026-10-07 13:20:42 +05:30

codex_headers.py

…

codex_responses_adapter.py

fix(auxiliary): let the shared normalizer judge tool-call-like aux text

2026-10-05 19:35:01 +05:30

codex_runtime_history_seed.py

…

codex_runtime.py

fix(codex): hermes chat -q codex exec/apply_patch approvals fail closed at once

2026-10-05 20:08:11 +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

fix(lint): stdin guard catches calls whose arguments start on the next line (review)

2026-10-07 01:03:03 +05:30

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_prellm.py

refactor(compaction): own the bench probe deadline load in PreLlmSkipMixin

2026-10-06 12:03:52 +05:30

context_compressor_summary.py

fix(compression): a keyless pinned route clears the main credential

2026-10-05 15:37:17 -07:00

context_compressor.py

refactor(compaction): name the pre-LLM summary skip in the fallback log

2026-10-06 12:03:52 +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_codex.py

refactor(compression): import the codex cooldown lookup at module level

2026-10-07 13:20:42 +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_telemetry.py

fix(telemetry): a host-timeout compression cancel counts as failed, not skipped

2026-10-05 16:18:47 -07:00

conversation_compression.py

refactor(compression): import the codex cooldown lookup at module level

2026-10-07 13:20:42 +05:30

conversation_loop.py

feat(voice): auxiliary.voice_chat routes spoken voice turns to their own model

2026-10-07 01:35:11 -07:00

copilot_acp_client.py

…

credential_persistence.py

…

credential_pool_admin.py

…

credential_pool_model_cooldowns.py

feat(desktop): a rate-limited provider says why and until when in the model pickers

2026-10-05 12:10:01 -05:00

credential_pool_plugin.py

feat(auth): brokered PKCE, expiry-driven rotation and aux on plugin OAuth rows

2026-10-05 23:33:41 -07:00

credential_pool.py

feat(auth): brokered PKCE, expiry-driven rotation and aux on plugin OAuth rows

2026-10-05 23:33:41 -07:00

credential_sources.py

…

credits_tracker.py

…

curator_backup.py

…

curator.py

fix(skills): background review can improve every skill, not just curator-managed ones

2026-10-07 02:27:40 -07:00

deadline.py

fix(lint): stdin guard catches calls whose arguments start on the next line (review)

2026-10-07 01:03:03 +05:30

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

fix(desktop): open the rejected provider's key from Update API key

2026-09-29 19:31:49 -05:00

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

feat(auth): brokered PKCE, expiry-driven rotation and aux on plugin OAuth rows

2026-10-05 23:33:41 -07:00

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

Inline agent tools get the shared argument repair (clarify, todo, memory, session_search) (#127781)

2026-09-29 22:07:30 +05:30

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

fix(kanban): retry a failed worker turn in place; carry the recovered response

2026-09-30 22:30:35 -04:00

lazy_forward.py

…

learn_prompt.py

…

learning_graph_render.py

…

learning_graph.py

fix(learning-graph): don't leak 'external' source to later skills in a root

2026-09-30 23:10:24 -05:00

learning_mutations.py

…

legacy_cli.py

…

lmstudio_reasoning.py

refactor(providers): one copy each of TokenHub's effort rule and the generic nested reasoning fallback

2026-10-07 11:27:35 +05:30

manual_compression_feedback.py

…

markdown_tables.py

…

memory_manager.py

fix(memory): scrub secrets on every memory-provider hand-off (#115104)

2026-10-07 02:06:15 -07:00

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

fix(compression): name the main model when aux summary falls back

2026-10-05 15:37:17 -07:00

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

refactor(compaction): drop a dead budget guard and read the user content once

2026-10-05 20:01:03 +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

feat(agent): truncated context files name the sections that were dropped

2026-10-04 23:05:42 -07:00

prompt_cache_boundary.py

…

prompt_cache_scope.py

Free tier: a rate limit ends a watched turn at once with the reset time (#134049)

2026-10-06 12:31:21 -07:00

prompt_caching.py

refactor(cache): one scaffold-split builder and one prior-decoration strip pass

2026-10-06 14:01:25 +05:30

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

refactor(providers): one copy each of TokenHub's effort rule and the generic nested reasoning fallback

2026-10-07 11:27:35 +05:30

reasoning_params.py

…

reasoning_promotion.py

refactor(agent): drop redundant str() wrapping around custom_provider_slug args

2026-10-06 13:01:39 +05:30

reasoning_summaries.py

…

reasoning_timeouts.py

…

redact.py

fix(memory): scrub secrets on every memory-provider hand-off (#115104)

2026-10-07 02:06:15 -07:00

relay_cwd.py

…

relay_llm.py

fix(relay): native Anthropic stream spans keep their input and cache token counts

2026-10-07 00:18:10 +05:30

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

Free tier: a rate limit ends a watched turn at once with the reset time (#134049)

2026-10-06 12:31:21 -07:00

review_engine.py

…

review_idle_queue.py

…

route_binding.py

feat(voice): auxiliary.voice_chat routes spoken voice turns to their own model

2026-10-07 01:35:11 -07:00

runtime_cwd.py

…

runtime_self_protection.py

…

sdk_transform_bypass.py

…

search_policy.py

fix(agent): a home working dir is a fallback, not a project

2026-09-29 14:53:00 -05:00

secret_scope.py

fix(secret-scope): compose the managed .env into every profile secret scope

2026-09-29 22:57:48 -04:00

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

…

session_source.py

Free tier: a rate limit ends a watched turn at once with the reset time (#134049)

2026-10-06 12:31:21 -07:00

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): tag the plugin-skill projection instead of invalidating from plugins.py

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

skill_preprocessing.py

fix(skills): trust-scope inline-shell auto-exec to non-community hub skills

2026-10-06 11:20:43 -04:00

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

docs(ssl): note SSL_CERT_FILE sits on top of the platform store

2026-10-06 12:25:01 +05:30

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

fix(agent): a home working dir is a fallback, not a project

2026-09-29 14:53:00 -05:00

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

refactor(security): remove the bundled tirith scanner from core

2026-10-06 14:22:13 -07:00

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

fix(tools): an image-bearing MCP result keeps its incomplete-data notice (review)

2026-10-06 22:54:13 -07:00

tool_executor.py

fix(tools): image-bearing tool results stay inside the turn budget; spills and refusals keep their content (review)

2026-10-06 22:54:13 -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

feat(stt): live transcription while you speak on CLI, TUI and Desktop (stt.streaming)

2026-10-06 12:50:31 -07:00

transcription_registry.py

…

tts_provider.py

fix(tts): validate a plugin's stream rate once; keep nous off plugin routing

2026-10-06 00:52:41 -07:00

tts_registry.py

…

turn_api_call.py

fix(desktop): mark replies the user stopped as "Response stopped"

2026-10-01 03:27:39 -05:00

turn_api_error.py

Free tier: a rate limit ends a watched turn at once with the reset time (#134049)

2026-10-06 12:31:21 -07:00

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): gate reasoning-only answer promotion by route

2026-10-06 13:01:39 +05:30

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

fix: use original gateway text for automatic session titles

2026-09-29 15:57:52 -05:00

turn_failure_copy.py

Free tier: a rate limit ends a watched turn at once with the reset time (#134049)

2026-10-06 12:31:21 -07:00

turn_final_response.py

refactor(agent): derive reasoning-promotion trust live in agent/reasoning_promotion

2026-10-06 13:01:39 +05:30

turn_finalizer.py

feat(voice): auxiliary.voice_chat routes spoken voice turns to their own model

2026-10-07 01:35:11 -07:00

turn_iteration_prep.py

Free tier: a rate limit ends a watched turn at once with the reset time (#134049)

2026-10-06 12:31:21 -07:00

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

Free tier: a rate limit ends a watched turn at once with the reset time (#134049)

2026-10-06 12:31:21 -07:00

turn_recovery.py

fix(agent): a credit-limited 402 retries once with the affordable output cap instead of failing as billing

2026-10-07 06:41:07 +05:30

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

fix(agent): a credit-limited 402 retries once with the affordable output cap instead of failing as billing

2026-10-07 06:41:07 +05:30

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

feat(voice): auxiliary.voice_chat routes spoken voice turns to their own model

2026-10-07 01:35:11 -07:00

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

fix(verify): share only a full passing hermes verify across sessions

2026-10-05 16:43:19 -05:00

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

…

voice_turn_route.py

feat(voice): voice turns default to reasoning off, the lowest valid level where off is refused

2026-10-07 01:35:11 -07:00

web_search_provider.py

…

web_search_registry.py

fix(web): pick own key or Nous Tool Gateway per web capability in Desktop

2026-10-06 02:53:56 -07:00