Sync tasks lacked notify — service ran stale code after rsync, causing 502
on /api/health and login failures. All three sync tasks now notify handlers.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(ansible): add sync-code-source.yml to push code from controller to SLM (#3604)

Adds a new playbook that rsync's the dev machine's local checkout to
/opt/autobot/code_source on the SLM manager.  Integrated into
deploy-slm-manager.yml as a no-op step when controller_repo_root is
not set, and active when passed via -e:

  ansible-playbook ... -e "controller_repo_root=/path/to/AutoBot-AI"

Closes the offline-deployment gap where the pre-flight GitHub pull
silently falls back to stale code_source, causing fixes that exist in
the local repo to never reach the SLM.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ansible): replace fragile group-based SLM detection in Phase 4c with filesystem stat

The _is_slm_manager check relied on inventory group membership
(slm_server/slm) or node_roles values, both of which are absent in
temp inventories generated by per-node provisioning from the SLM UI.
This caused Phase 4c to skip the nginx co-location re-render entirely,
leaving / → /slm/ redirect intact even when the user frontend was
deployed on the same host.

Replace with filesystem-based detection:
  1. Stat /etc/nginx/sites-available/autobot-slm to identify the SLM manager
  2. Stat autobot-frontend/package.json to confirm co-location

Both stats work regardless of inventory shape (wizard, per-node, static).
Also remove the role_path fallback in the template src — the playbook_dir
relative path is always correct and avoids a None-based string.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ansible): use set_fact to override slm_colocated_frontend — include_vars beats task vars

include_vars has Ansible precedence 18 while task-level vars: has 17.
Loading slm_manager/defaults/main.yml (which has slm_colocated_frontend: false)
via include_vars was silently overriding vars: slm_colocated_frontend: true
on the template task, so the template always rendered in non-co-located mode.

Replace vars: on the template task with an explicit set_fact (precedence 19)
that runs after include_vars, ensuring the template sees true.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ansible): fix code_source ownership before git reset to unblock pre-flight sync

Earlier sudo cp commands left provision-fleet-roles.yml owned by root:root
in code_source, causing git reset --hard to fail with EPERM. Add a
pre-flight chown (become: true) that runs before the git pull so force: yes
can always overwrite any root-owned file regardless of how it got there.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ansible): preserve wizard's slm_colocated_frontend in role set_fact; fix include_vars precedence in Phase 4c

slm_manager role set_fact was unconditionally overwriting slm_colocated_frontend
with the file-detection result, discarding the True value the wizard set in the
inventory.  Change to OR expression so a wizard-supplied True is preserved.

Phase 4c had include_vars (precedence 18) loading defaults with false, then
task vars: (precedence 17) trying to set true — always losing.  Add an explicit
set_fact after include_vars so the correct value wins.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ansible): restart backend+celery when code is synced (#3608)

Sync tasks lacked notify — service ran stale code after rsync, causing 502
on /api/health and login failures. All three sync tasks now notify handlers.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ansible): correct systemd startup ordering for all core services (#3609)

- autobot-backend: redis.service → redis-stack-server.service (was silently
  ignored by systemd); add Wants= + network-online.target
- autobot-celery: same redis fix; keep After=autobot-backend.service
- autobot-slm-backend: add redis-stack-server + postgresql dependencies
  (had none — could race both on every boot)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(rag): per-user annotation signals for personalized RAG retrieval (#3240) (#3610)

* feat(rag): per-user annotation signals for personalized RAG retrieval (#3240)

* fix(rag): export GLOBAL_USER publicly, fix query_text field, Literal validation, move imports (#3240)

* fix(backend): replace naive datetime.now() with UTC-aware calls (#3613) (#3615)

* fix(backend): replace naive datetime.now() with UTC-aware calls throughout (#3613)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend): fix aware/naive datetime comparisons in agent_client + auth_middleware (#3613)

- agent_client.py: default sentinel for last_health_check was datetime.min (naive);
  cold-start subtraction from datetime.now(tz=utc) raised TypeError — use
  datetime.min.replace(tzinfo=timezone.utc) instead
- auth_middleware.py: fromisoformat() of pre-migration Redis strings returns naive;
  normalize to UTC before comparing against now(tz=utc)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(memory-graph): consolidate into single autobot_memory_graph with semantic search (#3612) (#3616)

* feat(memory-graph): add semantic search and hybrid scoring to autobot_memory_graph (#3612)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(memory-graph): fix _parse_ft_results, SECRET_USAGE pattern, ssot_config.get, asyncio.run (#3612)

- _parse_ft_results now extracts keys and _redis_search calls _fetch_entities_by_keys
- Add SECRET_USAGE to secret/credential pattern in _ENTITY_TYPE_PATTERNS
- Replace ssot_config.get() with getattr() (Pydantic model, not dict)
- Replace asyncio.get_event_loop().run_until_complete() with asyncio.run() in all 6 test sites

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(constants): add TTL_1_HOUR/TTL_5_MINUTES and replace raw cache TTL literals (#3614) (#3617)

* refactor(constants): add TTL_1_HOUR/TTL_5_MINUTES and replace raw cache TTL literals (#3614)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(constants): fix orphan TTL import, residual literals, noqa suppressions in #3617

- router.py: replace raw 300 with TTL_5_MINUTES (resolves orphan import)
- advanced_cache_manager.py: add TTL_1_HOUR import, replace all 8 raw 300/3600 literals
- embedding_cache.py: remove incorrect # noqa: F401 comment
- token_optimizer.py: remove incorrect # noqa: F401 comment

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(slm): add admin password reset to user management UI (#3625)

* feat(slm): add admin password reset to user management UI (#3625)

- POST /slm-users/{id}/change-password and /autobot-users/{id}/change-password
  backend endpoints using UserService.change_password(require_current=False)
- changeSlmUserPassword() + changeAutobotUserPassword() in useSlmUserApi.ts
- PasswordChangeForm: apiEndpoint prop replaces hardcoded /api/users/ URL
- UserManagementSettings: key (lock) icon in each user row opens reset modal;
  selectedUserType tracks slm/autobot/legacy to route to correct endpoint

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(slm): remove unreachable InvalidCredentialsError handler; use getSlmApiBase() for password change URLs (#3625)

- Drop except InvalidCredentialsError blocks — require_current=False means the
  service never raises it; dead code flagged in code review
- passwordChangeApiEndpoint now uses getSlmApiBase() from ssot-config instead
  of hardcoded /api/ prefix, fixing co-located /slm/api deployments

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): add getApiBase() to ssot-config for configurable API prefix (#3628)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend): use local time in is_business_hours and is_weekend (#3619)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend): normalize naive datetime from Redis in knowledge_manager (#3618)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend): normalize naive datetimes from Redis in stats and analytics (#3620)

Closes #3620

Pre-#3615 Redis data stores naive datetime strings. After fromisoformat()
parsing, add UTC tzinfo when tzinfo is None to prevent TypeError when
comparing with UTC-aware datetime.now(tz=timezone.utc).

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(slm-frontend): replace hardcoded /api/ with getSlmApiBase() in 3 remaining files (#3627)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): adopt getApiBase() in composables (#3630)

- Added getApiBase() to ssot-config.ts (prerequisite from #3628)
- Replaced ~133 hardcoded '/api/' occurrences across 36 composable files
- All replacements use template literals: `${getApiBase()}/path`
- Imports added or extended in each modified file

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): adopt getApiBase() in service and utils layer (#3629)

* chore(frontend): adopt getApiBase() in service and utils layer (#3629)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): remove duplicate getApiBase() — defined in #3632 (#3629)

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): adopt getApiBase() in stores, models, and components (#3631) (#3641)

- Added getApiBase() to ssot-config.ts (prerequisite from #3628)
- Replaced 110 hardcoded /api/ occurrences across 8 files:
  - stores/useKnowledgeStore.ts (7 paths)
  - stores/usePermissionStore.ts (10 paths)
  - stores/useUserStore.ts (2 paths)
  - models/repositories/KnowledgeRepository.ts (36 paths)
  - models/repositories/ChatRepository.ts (12 paths)
  - models/repositories/SystemRepository.ts (37 paths)
  - models/repositories/ApiRepository.ts (5 paths incl. cache keys)
  - components/ChatInterface.ts (1 path)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* test: replace asyncio.get_event_loop() with asyncio.run() (#3605)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(helpers): add extra_data support to TaskResult (#3564)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(helpers): remove duplicate raise_not_found_error in catalog_http_exceptions (#3566)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(helpers): cap exponential backoff via TimingConstants (#3565)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(backend): remove dead AgentThresholds import and replace magic 0.8 in orchestrator.py (#3607)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(agents): declare _get_system_prompt @abstractmethod in StandardizedAgent (#3606)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(ci): align isort line_length with Black at 120 in pyproject.toml (#3408)

Align pyproject.toml isort line_length from 100 to 120 to match Black's
line-length and the explicit --line-length=120 args in pre-commit and CI.
Black and isort hooks were already present in both pre-commit-config.yaml
and code-quality.yml; this fix resolves the config inconsistency that
caused isort --settings-path=. to use a conflicting line_length value.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ansible): sync standalone autobot_shared on backend deploy (#3649)

The backend role only synced code_source/autobot_shared/ into
backend_install_dir/autobot_shared/ (the in-backend copy). The separate
/opt/autobot/autobot_shared/ directory — resolved via PYTHONPATH by
Celery workers and other services — was never updated, causing
ModuleNotFoundError for newly added modules (pagination.py,
task_result.py, error_boundaries.py, alert_cooldown.py).

Add a synchronize task immediately after the existing backend sync that
also pushes to the standalone path, plus a file ownership fix task.
Introduce backend_shared_standalone_dir default (/opt/autobot/autobot_shared)
so the path is a named variable, not a hardcoded literal.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): remove duplicate getApiBase() declarations from ssot-config.ts (#3631)

All target files (stores, model repositories, ChatInterface) already had
getApiBase() imported and in use from prior development. Fixed the only
remaining issue: three duplicate getApiBase() function declarations in
ssot-config.ts (added by parallel PRs #3628-#3630) — reduced to a single
canonical declaration.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(tests): replace hardcoded localhost:8001 with get_test_backend_url() helper

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(slm-frontend): replace hardcoded /autobot-api/ with getBackendUrl()

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend): guard context_window_manager against list-format llm_models.yaml (#3647)

PR #3588 changed models: to a list-based registry; ContextWindowManager
still accessed it as a dict, crashing all workers on startup.
Detect list format and fall back to default context-window config.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(slm): pass auth token to PasswordChangeForm — was always sending empty Bearer

PasswordChangeForm read localStorage.getItem('authToken') which is never set;
SLM auth store uses key 'slm_access_token' in sessionStorage.
Added authToken prop; UserManagementSettings passes authStore.token.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ansible): guard groups['database'] lookup in deploy-backend-local.yml (#3651)

Wrap the hostvars lookup with an inline if-guard so AnsibleUndefinedVariable
is never raised when the 'database' inventory group is absent or empty.
Falls back to 127.0.0.1 in both cases.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend): default SLM_URL to localhost and deduplicate startup warning (#3655)

On co-located deployments SLM_URL is not set, so the backend now defaults
to http://127.0.0.1:8000 instead of None, preventing silent SLM client
failures.  The startup warning is deferred to init_slm_client() and gated
by a module-level flag, so it fires at most once per process instead of 3×.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(backend): give context_window_manager its own config YAML (#3650)

Decouple ContextWindowManager from llm_models.yaml (which uses a list
schema for the YAML-model-registry feature) by introducing a dedicated
config/context_windows.yaml with the dict-keyed schema the manager has
always expected.  Covers all 40+ models from llm_models.yaml with
realistic context_window_tokens, max_output_tokens, and message_budget
values.  Default config path updated from config/llm_models.yaml to
config/context_windows.yaml.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend): add TTL_24_HOURS to chat_workflow/manager.py import (#3646)

PR #3617 added TTL_24_HOURS usage in manager.py but the import from
constants.ttl_constants was never updated — NameError crashes all workers.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(agents): cap exponential backoff in agent_client.py via exponential_backoff_delay (#3658)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(backend): remove dead doc_generation_threshold attribute (#3656)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend): fix REDIS_HOST → AUTOBOT_REDIS_HOST in chat history modules

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): remove duplicate getApiBase() declarations from ssot-config.ts (#3628)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(slm-frontend): replace hardcoded /api/ with getSlmApiBase() (#3627) (#3676)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(ansible): add pre-flight code_source sync to deploy playbooks (#3604)

Create shared pre-flight-code-sync.yml extracted from provision-fleet-roles.yml
Play 0; import it at the top of deploy-full.yml, deploy-aiml.yml,
deploy-backend-local.yml, and deploy-backend-remote.yml so standalone
deployments always pull the latest code before roles run.  GitHub sync uses
ignore_errors so air-gapped SLMs fall back to existing code_source with a
visible warning and staleness report.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(slm): add component selector to File Drift Check UI (#3433)

Add a <select> dropdown bound to selectedDriftComponent (default: autobot-slm-backend)
in the File Drift Check card. The selected value is passed to fetchDrift() so the
API call includes the correct ?component= query param. A watcher clears stale drift
results when the user switches components.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(ci): add doc linting to catch stale file references in developer guides (#3425)

- Add pipeline-scripts/check-doc-references.py that validates backtick-quoted
  file path references in the 9 canonical developer docs resolve to existing
  files; searches by full path then by filename across all source trees.
- Wire the script as a new step in code-quality.yml before the summary step.
- Fix 13 stale references in AUTOBOT_REFERENCE, DEVELOPER_SETUP,
  REDIS_CLIENT_USAGE, LOGGING_STANDARDS, and ERROR_CODE_CONVENTIONS that the
  new linter caught: removed deleted redis_database_manager.py entries, updated
  tls.yml → nginx/tasks/main.yml, corrected troubleshooting/security doc paths,
  updated logging config path, and replaced non-existent error_migration_map.yaml
  and test_error_catalog.py references with current file locations.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): adopt getApiBase() in Vue components and views (Phase 4)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(analytics): scope codebase analytics caches and live scans to source_id (#3685)

Five cross-project data leakage vectors fixed:
1. _duplicate_cache keyed by source_id (was a single Optional[dict])
2. _ownership_analysis_cache keyed by source_id (same pattern)
3. Live duplicate analysis resolves source.clone_path so DuplicateCodeDetector
   scans the correct project instead of always the AutoBot repo
4. detect_config_duplicates_endpoint adds source_id param + clone_path resolution
5. indexing_tasks entries now store source_id; _get_active_indexing_task filters
   by it so Project B stats don't show Project A's indexing progress

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): adopt getApiBase() in chat Vue components (Phase 5b)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend): replace bare REDIS_HOST with AUTOBOT_REDIS_HOST in chat_history (no-op — already in #3672)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(slm-frontend): replace hardcoded /autobot-api/ with getBackendUrl() (#3652) (no-op — already in #3662)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(slm-frontend): safe WebSocket URL construction when getBackendUrl() is absolute (#3673)

Add buildWsUrl() helper that strips the http(s):// scheme from getBackendUrl()
and replaces it with the correct ws(s):// scheme. Handles proxy mode (empty
base URL) by falling back to window.location.host. Fixes the broken URL
produced by prepending ws:// to an already-absolute URL.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(infrastructure): standardise Redis file ownership to autobot:autobot (#3396)

- Add redis_owner/redis_group defaults (both "autobot") to redis role
- Add idempotent file tasks to create and chown /var/lib/redis,
  /etc/redis, /var/run/redis, /var/log/redis with recurse for data
  and config dirs; tagged 'redis,ownership' for selective runs
- Update systemd service override to run redis-stack-server as
  autobot:autobot instead of redis:redis so all written files
  inherit the correct ownership automatically

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): use getApiBase() instead of getBackendUrl() in notification composables (#3675)

Replace getBackendUrl() with getApiBase() in useNotificationConfig (lines 72, 94)
and remove the getBackendUrl() prefix from apiRequest in useWorkflowNotificationConfig
(line 80) so all notification API calls use the canonical /api path prefix.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): replace hardcoded /api/advanced/ with getApiBase() in BusinessIntelligenceView (#3674)

All 5 hardcoded /api/advanced/ calls were replaced with ${getApiBase()}/advanced/...
template literals in BusinessIntelligenceView.vue. Import added at line 320.
Fix was included in Phase 4 commit 2aca9bc08 — this commit closes the issue.

Closes #3674

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): adopt getApiBase() in knowledge Vue components (#3682) (#3703)

Replace all hardcoded '/api/' path prefixes with `${getApiBase()}/`
in 20 knowledge components. Add `import { getApiBase } from
'@/config/ssot-config'` to each file. Covers single-quoted strings
and template literals across apiClient and apiService call sites.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): adopt getApiBase() in security, visualizations, workflow, and research Vue components (#3684) (#3706)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): adopt getApiBase() in knowledge Vue components (Phase 5c) (#3629)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(analytics): scope quality/review/evolution/generation to project source_root (#3441)

- analytics_shared: add resolve_source_root_or_404() returning Path so callers
  can scope query results, not just validate the source exists
- analytics_quality: _get_problems_from_chromadb() and
  calculate_real_quality_metrics() accept source_root; problems whose
  file_path does not resolve under source_root are excluded; Redis cache key
  is namespaced per source_root so per-project caches are independent
- analytics_code_review: analyze_diff() resolves source_root and skips any
  file whose resolved path falls outside source_root
- analytics_evolution: _fetch_timeline_snapshots() and
  _fetch_trend_snapshots_sync() accept an evolution_prefix parameter;
  _build_evolution_prefixes() builds the evolution:{source_id}: namespace;
  timeline, patterns, and trends endpoints use the scoped prefix so each
  project reads/writes its own Redis keys; added
  _extract_pattern_types_from_prefix() to generalise the existing helper
- analytics_code_generation: CodeGenerationEngine.get_stats() accepts
  source_id and namespaces its Redis key as stats:{source_id}:{date} when
  provided; the /stats endpoint passes source_id through

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(frontend): correct useEvolution base URL from /analytics/evolution to /evolution (#3698)

* chore(frontend): adopt getApiBase() in analytics Vue components (#3681)

Replace all hardcoded '/api/' strings with getApiBase() across 12
analytics Vue components. Add import { getApiBase } from
'@/config/ssot-config' to each file and update all fetchWithAuth and
api service call sites.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(backend): add explicit encoding='utf-8' to file I/O (#3391)

- Add encoding='utf-8' to open() call in event_manager.py
- Add media_type='application/json; charset=utf-8' to all JSONResponse
  calls in api/chat.py (11 sites)
- Add encoding='utf-8' to open() calls in code_analysis scripts and tools
  (analyze_env_vars, analyze_code_quality, analyze_architecture,
  analyze_frontend, generate_patches, code_quality_dashboard,
  logging_standardizer, setup.py)
- Skips binary-mode opens (rb/wb) and non-text opens (tarfile, Image)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(tests): co-locate single-module tests from autobot-backend/ root (#3364)

Move 5 unit tests that each test a single module into the directory of
the module they cover, following the co-location pattern established in
chat_workflow/ and other packages.

Moves:
- chat_intent_detector_test.py → chat_workflow/
- tool_discovery_test.py       → tools/
- encryption_service_test.py   → security/
- auth_rbac_test.py            → security/
- memory_package_test.py       → memory/

Tests in chat_workflow/ and tools/ confirmed passing after move.
Tests in security/ have a pre-existing collection error (structlog not
installed in local dev venv) that existed before this PR.
Tests in memory/ have a pre-existing collection error (aiosqlite not
installed in local dev venv) that existed before this PR.

E2E, integration, and cross-module tests remain at autobot-backend/ root.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(frontend): remove spurious /analytics prefix from code-review pattern preference URLs (#3699)

* fix(frontend): use wss:// on HTTPS for CodeQualityDashboard WebSocket connection (#3702)

* fix(frontend): remove dead loadReview() — GET /review/{id} has no backend storage (#3701)

* fix(ansible): AUTOBOT_CHROMADB_HOST fallback uses backend_ai_stack_host (#3541)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(slm): implement autobot-admin CLI with reset-password subcommand (#3691)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ansible): add post-deploy smoke test to backend role (#3687)

Wait for uvicorn to accept TCP connections on port 8001 and verify
/api/health returns 200 after every backend role execution.  Adds
backend_health_check_port/timeout defaults so values are configurable.
Also adds equivalent checks to update-node.yml Play 3 when
autobot-backend is among the deployed roles.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend): rate-limit AI Stack connection error logging (#3686)

Add _log_connection_error() helper with a 60-second suppression window: first
failure per window is emitted at WARNING, subsequent retries are demoted to
DEBUG so backend-error.log is not flooded when AI Stack is unavailable. Also
downgrade HTTP-error and unexpected-exception log calls from ERROR to WARNING
since these are expected transient conditions when the remote service is down.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ansible): idempotent autobot:autobot ownership on Redis dirs (#3396)

The command module with changed_when:true always reported "changed" and
would trigger restart redis-stack on every play run. Replaced with the
file module (recurse:yes) which only marks changed when ownership
actually differs, making the task fully idempotent. Also removed the
duplicate directory-creation task that preceded it.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(frontend): remove duplicate withSourceId() wrap in Redis health URL (#3714)

* fix(tests): update stale src.tool_discovery mock path (#3722)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(dev): add requirements-dev.txt with structlog and aiosqlite (#3723)

Creates requirements-dev.txt at repo root so local dev venvs include
structlog and aiosqlite, preventing ModuleNotFoundError during pytest
collection of security/ and memory/ test packages.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(frontend): wrap router-view with ErrorBoundary in App.vue (#3375)

Wire the existing ErrorBoundary.vue into App.vue by importing it,
registering it in the components map, and wrapping <router-view> so
any unhandled runtime error in a child view shows a user-friendly
fallback instead of a white screen.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(backend): persist code-review results in Redis + GET /review/{id} endpoint (#3716)

- Store each analyze result under code_review:result:{source}:{uuid} with TTL_7_DAYS
- Push summary entry to code_review:history:{source} list (capped at 100)
- Add GET /review/{review_id} endpoint for drill-down lookup
- GET /history now reads from Redis instead of returning no-data stub
- Frontend re-adds loadReview(id) and @click handler on history items
- Add reviewLoaded i18n key to all 11 locale files

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(slm): Redis service management UI — start/stop/restart controls (#3381)

Adds RedisServicePanel.vue with start/stop/restart controls, live status
polling every 10 s (uptime, memory, client count), and a stop confirmation
dialog. Wires the panel into ServicesView.vue below the summary cards.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(dev): uncomment aiosqlite in autobot-backend/requirements.txt (#3723) (#3745)

Makes aiosqlite an explicit dependency rather than relying on transitive
inclusion from ../requirements.txt. Required for security/ and memory/
test collection under pytest.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(analytics): fix flake8 E501 line-length violation in analytics_evolution.py (#3724) (#3741)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(tests): correct stale src.tool_discovery patch path in tool_discovery_test (#3722) (#3738)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend): remove total timeout from LLM streaming — use connect-only timeout (#3732) (#3737)

Replace ClientTimeout(total=TIMEOUT_HTTP_DEFAULT) with ClientTimeout(total=None,
connect=TIMEOUT_HTTP_DEFAULT) so the 60s cap no longer kills long-running LLM
generation. Connection failures still time out fast via the connect param.

Also improve the user-facing error message for TimeoutError from the generic
"{type}: unexpected error" to "The model is taking too long to respond. Please
try again."

Closes #3732

* chore(tests): extend get_test_backend_url() to remaining 14 backend test files (#3661) (#3727)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(slm-frontend): replace remaining 2 hardcoded /api/ with getSlmApiBase() (#3725) (#3733)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(browser): replace hardcoded :3000 with NetworkConstants.BROWSER_SERVICE_PORT in web_crawler.py fallback (#3728) (#3731)

* fix(browser): replace hardcoded :3000 with NetworkConstants.BROWSER_SERVICE_PORT in web_crawler.py fallback (#3728)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ansible): inject AUTOBOT_BROWSER_SERVICE_HOST into backend.env from fleet vars (#3728)

backend.env.j2 never set AUTOBOT_BROWSER_SERVICE_HOST, so PlaywrightService
raised ValueError on every Ansible-provisioned deployment and BROWSER_VM_IP
defaulted to empty string causing health checks to hit http://:3000/health.

- backend.env.j2: render AUTOBOT_BROWSER_SERVICE_HOST/AUTOBOT_BROWSER_HOST
  from browser_host (fleet wizard) or backend_browser_host (direct playbooks)
- defaults/main.yml: add backend_browser_host="" and backend_browser_port=3000
- deploy-backend-remote.yml / deploy-backend-local.yml: resolve browser host
  from groups['browser'][0] inventory, same pattern as redis/ai_stack hosts

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): adopt getApiBase() in JS utility files (#3726) (#3730)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(tests): evaluate get_test_backend_url() lazily in TakeoverTestClient (#3747) (#3756)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(frontend): adopt getApiBase() in remaining 7 SecretsApiClient.js paths (#3746) (#3757)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(tests): handle https→wss protocol upgrade in simple_terminal.e2e_test.py

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(frontend): clear auto-save interval on ChatController destroy

* fix(frontend): guard loadReview against undefined reviewId (#3755) (#3759)

Pre-existing Redis history entries (created before PR #3735) have no id
field. Clicking them called loadReview(undefined), which sent
GET /review/undefined and produced a confusing 404 error toast.

Replace the silent early-return with showToast(reviewNotAvailable,
'warning') so the user sees a clear message instead. Added the
reviewNotAvailable i18n key to all 11 locale files.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(config): replace hardcoded 172.16.168.* IPs with named placeholders in .env.example

* fix(monitoring): replace hardcoded 172.16.168.25 with $browser_vm_ip variable in Grafana dashboard

* fix(config): canonicalize browser host env var to AUTOBOT_BROWSER_SERVICE_HOST

* fix(frontend): clear auto-save interval on ChatController destroy (#3749) (#3766)

* chore(frontend): adopt getApiBase() in ApiClientMonitor.js (#3752) (#3767)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(chat-workflow): route system prompt via Ollama system field; raise num_ctx to 8192 (#3761) (#3765)

* chore(frontend): adopt getApiBase() in AppConfig.js and ServiceDiscovery.js (#3751) (#3769)

Replace 5 hardcoded '/api/' path literals with getApiBase() calls in
AppConfig.js (lines 382, 475, 559) and ServiceDiscovery.js (lines 403, 429).

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend): allow any authenticated user to GET /review/{id} (#3754) (#3772)

Replace check_admin_permission with get_current_user on get_review_by_id
so regular users can drill into their own code-review history entries.
All other write/admin endpoints retain check_admin_permission.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ansible): inject SLM secrets into standalone deploy extra_vars (#3773)

* fix(ansible): inject SLM secrets into standalone deploy extra_vars (#3519)

* fix: guard against None variables in _run_playbook merge

* feat(frontend): add KB vectorization status badge, action button, batch toolbar, progress modal (#3388) (#3774)

* perf(autoresearch): replace HumanReviewScorer polling loop with BLPOP (#3781)

* fix(ansible): inject SLM secrets into standalone deploy extra_vars (#3519)

* fix: guard against None variables in _run_playbook merge

* perf(autoresearch): replace HumanReviewScorer polling loop with BLPOP (#3209)

* perf(backend): convert file I/O to aiofiles in chat_history_manager and simple_pty (#3783)

* fix(ansible): inject SLM secrets into standalone deploy extra_vars (#3519)

* fix: guard against None variables in _run_playbook merge

* perf(backend): convert file I/O to aiofiles in chat_history_manager and simple_pty (#3392)

* chore(frontend): add i18n keys for KB vectorization UI components (#3785)

* fix(ansible): inject SLM secrets into standalone deploy extra_vars (#3519)

* fix: guard against None variables in _run_playbook merge

* chore(frontend): add i18n keys for KB vectorization UI components (#3779)

* fix(ansible): inject SLM secrets into standalone deploy extra_vars (#3519) (#3780)

* fix(ansible): inject SLM secrets into standalone deploy extra_vars (#3519)

* fix: guard against None variables in _run_playbook merge

* feat(memory): Redis-backed session-scoped working memory with TTL (#3775)

* feat(memory): Redis-backed session-scoped working memory with TTL (#3768)

- Add WorkingMemoryService with store/get/list/clear async methods
- Key pattern: autobot:session:{session_id}:memory:{key} in 'knowledge' DB
- Register singleton via UnifiedMemoryManager.working_memory property
- Add TTL_WORKING_MEMORY_DEFAULT = 3600 to ttl_constants.py
- Unit tests: 9 cases covering all methods + manager integration

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(memory): connection caching, error handling, race-free singleton (#3768)

- WorkingMemoryService: lazy-init _redis instance attribute via _get_redis()
  so the Redis client is created once per service instance, not per call
- All four public methods wrapped in try/except that logs a warning and re-raises
- UnifiedMemoryManager: _working_memory initialized eagerly in __init__ to
  eliminate the race condition in the lazy property

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(memory): context-adaptive memory compression for small-context models (#3776)

* feat(memory): context-adaptive memory compression for small models (#3770)

Add ContextCompressionService that summarises dropped history and
re-ranks KB results by score to fit within model token budgets.
Wire into ContextWindowManager via async_should_compress().
Add compression_threshold field to every entry in context_windows.yaml.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(memory): compression guard logic, role:system mid-history, phi3 threshold, wire to llm_handler (#3770)

- Fix phi3 compression_threshold from 8192 to 4096 (matches context_window_tokens)
- Remove large-model early-return guard from should_compress; YAML thresholds are the sole policy
- Change summary role from 'system' to 'assistant' to avoid mid-history system messages
- Add model_thresholds constructor param to ContextCompressionService to avoid double YAML load
- Wire compress_kb_results into _prepare_llm_request_params after KB retrieval
- Update tests to reflect corrected behavior

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(chat-workflow): remove system_prompt from _build_continuation_prompt to fix double-injection (#3784) (#3791)

* chore(chat-workflow): remove orphaned system_prompt param from _build_full_prompt (#3794) (#3801)

* feat(agents): enforce memory read/write via pipeline lifecycle hooks (#3777)

* feat(agents): _before_process/_after_process memory lifecycle hooks (#3771)

Add async _before_process / _after_process hooks to StandardizedAgent.process_request()
so agents can optionally read from and write to WorkingMemoryService (issue #3768)
without breaking any existing agent when the dependency is absent.

- WorkingMemoryService imported with try/except ImportError fallback to no-op stubs
- Default hook implementations are no-ops — all 24+ existing agents unchanged
- Hook exceptions are caught and logged at WARNING; never re-raised
- Hook timing logged at DEBUG level (before/after elapsed ms)
- SentimentAnalysisAgent demonstrates the override pattern
- 9 unit tests verify call order and failure isolation

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(agents): correct WMS method names, remove private flag import, add pass to _after_process (#3771)

- Replace WorkingMemoryService.load/save() static calls with instance
  get()/store() via self._wm stored in __init__
- Remove fragile _working_memory_available import from standardized_agent;
  use local try/except with _WMS alias in sentiment_analysis_agent
- Add pass to empty _after_process base body in standardized_agent
- Document enriched context availability in process_request comment
- Fix 3 pre-existing E501 violations in standardized_agent (line wraps)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix: correct WorkingMemoryService import path (autobot_backend. → memory.)

* fix(agents): correct memory import path in standardized_agent (#3771)

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(ansible): move _SECRET_TO_ANSIBLE_VAR to shared location to avoid drift (#3778) (#3782)

Extract the duplicated secret-key -> Ansible-variable mapping from
setup_wizard.py and playbook_executor.py into a new shared module
services/ansible_secrets.py. Both callers now import from the single
source of truth, eliminating the risk of the two copies drifting apart.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(backend): expand hardcoding prevention for file paths and model names (#3397) (#3799)

- Add PathConfig to ssot_config with AUTOBOT_BASE_DIR/AUTOBOT_PLUGINS_DIR/etc fields
- Replace hardcoded /opt/autobot/plugins/* in plugin_manager.py with config.path.plugins_path
- Replace hardcoded /opt/autobot/docs in mcp_manual_integration.py with AUTOBOT_BASE_DIR env var
- Replace hardcoded "llama3.2:latest" in rlm/types.py, agent_loop/think_tool.py,
  rlm/benchmark.py with ROUTING_MODEL from ssot_config
- Add check_hardcoded_paths() and check_hardcoded_model_names() to pre-commit hook
- Exclude constants definition files (path_constants.py etc.) from hook scanning

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(ansible): add deploy-hybrid-docker.yml and docker Ansible role (#3424) (#3786)

* feat(ansible): add deploy-hybrid-docker.yml playbook and docker role (#3424)

Create the missing Ansible playbook and role that
SLMDeploymentOrchestrator.deploy_docker() targets. Without these files
every call to POST /api/slm/deployments/docker fails immediately with
FileNotFoundError from PlaybookExecutor.execute_playbook().

- autobot-slm-backend/ansible/deploy-hybrid-docker.yml: standalone
  playbook that runs the docker role; accepts target_host and
  docker_containers extra_vars passed by the orchestrator
- autobot-slm-backend/ansible/roles/docker/tasks/main.yml: installs
  Docker Engine (Debian + RedHat), starts the service, adds the service
  user to the docker group, pulls images and runs containers from the
  docker_containers list, then asserts all containers reach running state
- autobot-slm-backend/ansible/roles/docker/defaults/main.yml: defaults
  for all vars referenced in the role (docker_image, docker_containers,
  docker_default_restart_policy, docker_verify_timeout, etc.)
- autobot-slm-backend/ansible/roles/docker/handlers/main.yml: restart
  docker handler
- autobot-slm-backend/ansible/deploy.yml: add docker branch so the
  standard DeploymentService path also resolves the docker role when
  target_roles includes "docker"

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix: route deploy.yml override from extra_data; add docker to DEFAULT_ROLES; fix arch default

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(knowledge): semantic duplicate guard on individual fact writes (#3788) (#3800)

Add FactsMixin._find_duplicate() that queries ChromaDB for near-duplicate
content before every store_fact() insert. Threshold (default 0.92) is
configurable via AUTOBOT_KB_DEDUP_THRESHOLD / config.cache.l2.kb_dedup_threshold.
Nine unit tests cover above-threshold skip, below-threshold write, exact
hash dedup, graceful ChromaDB error handling, and empty-KB no-false-positive.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(agents): per-agent cross-session diary — persistent agent journal in KB (#3792)

* feat(agents): per-agent cross-session diary in KB (#3789)

Add AgentDiaryService that stores timestamped journal entries as KB
facts (category AGENT_DIARY, source=agent_name) enabling cross-session
memory and retrospective semantic search per agent.  Wire diary write
into SentimentAnalysisAgent.process_query() as a concrete usage example.
10 unit tests — all passing.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(agents): diary read() uses metadata filter instead of semantic search (#3789)

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(memory): temporal fact validity — valid_from/valid_to on memory graph entities and relations (#3798)

* feat(memory): temporal fact validity — valid_from/valid_to on memory graph (#3790)

- Add valid_from/valid_to to entity metadata in _prepare_entity_metadata()
- Add valid_from/valid_to to both relation builder helpers
- Add EntityOperationsMixin.invalidate_entity() — marks entity expired without deleting
- Add RelationOperationsMixin.invalidate_relation() — marks a specific edge expired
- Add include_expired param to search_entities() and _fallback_search() (default False)
- Add QueryOperationsMixin.get_entities_as_of() — point-in-time entity query
- Legacy entities without valid_to are treated as currently valid (no migration needed)
- 21 new tests, all passing

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(memory): _is_entity_valid: future valid_to is still valid (#3790)

An entity with valid_to set to a future timestamp is still currently
valid. Only treat valid_to as expired when the timestamp is in the past.
Add datetime/timezone import; update test assertion to match.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(memory): essential story layer — always-loaded compact memory summary (#3793)

* feat(memory): essential story layer — always-loaded compact memory summary (#3787)

- Add EssentialStoryGenerator in memory/essential_story.py: fetches top
  facts by quality_score, respects per-model token budget, caches result
  in Redis knowledge DB with TTL_5_MINUTES, never raises from generate()
- Add essential_story_tokens to every model entry in context_windows.yaml
  (300 for ≤8192, 600 for ≤32768, 800 for >32768 context windows)
- Inject story into _prepare_llm_request_params in llm_handler.py after
  _get_system_prompt() so every LLM call carries persistent top-memories
- Add 18 unit tests covering generation, token budget enforcement,
  cache hit/miss, empty KB, and never-raise guarantee

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(memory): cap essential story KB fetch at top 50 facts (#3787)

Sort all facts by quality_score descending, then slice to 50 before the
token-budget loop. Avoids processing the entire KB on every cache miss.
Added test_caps_at_50_facts_before_token_loop to verify the hard cap.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(knowledge): semantic duplicate guard on individual fact writes (#3788) (#3805)

Add FactsMixin._find_duplicate() that queries ChromaDB for near-duplicate
content before every store_fact() insert. Threshold (default 0.92) is
configurable via AUTOBOT_KB_DEDUP_THRESHOLD / config.cache.l2.kb_dedup_threshold.
Nine unit tests cover above-threshold skip, below-threshold write, exact
hash dedup, graceful ChromaDB error handling, and empty-KB no-false-positive.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(memory): essential story layer — always-loaded compact memory summary (#3787) (#3804)

* feat(memory): essential story layer — always-loaded compact memory summary (#3787)

- Add EssentialStoryGenerator in memory/essential_story.py: fetches top
  facts by quality_score, respects per-model token budget, caches result
  in Redis knowledge DB with TTL_5_MINUTES, never raises from generate()
- Add essential_story_tokens to every model entry in context_windows.yaml
  (300 for ≤8192, 600 for ≤32768, 800 for >32768 context windows)
- Inject story into _prepare_llm_request_params in llm_handler.py after
  _get_system_prompt() so every LLM call carries persistent top-memories
- Add 18 unit tests covering generation, token budget enforcement,
  cache hit/miss, empty KB, and never-raise guarantee

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(memory): cap essential story KB fetch at top 50 facts (#3787)

Sort all facts by quality_score descending, then slice to 50 before the
token-budget loop. Avoids processing the entire KB on every cache miss.
Added test_caps_at_50_facts_before_token_loop to verify the hard cap.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(memory): temporal fact validity — valid_from/valid_to on memory graph (#3790) (#3807)

* feat(memory): temporal fact validity — valid_from/valid_to on memory graph (#3790)

- Add valid_from/valid_to to entity metadata in _prepare_entity_metadata()
- Add valid_from/valid_to to both relation builder helpers
- Add EntityOperationsMixin.invalidate_entity() — marks entity expired without deleting
- Add RelationOperationsMixin.invalidate_relation() — marks a specific edge expired
- Add include_expired param to search_entities() and _fallback_search() (default False)
- Add QueryOperationsMixin.get_entities_as_of() — point-in-time entity query
- Legacy entities without valid_to are treated as currently valid (no migration needed)
- 21 new tests, all passing

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(memory): _is_entity_valid: future valid_to is still valid (#3790)

An entity with valid_to set to a future timestamp is still currently
valid. Only treat valid_to as expired when the timestamp is in the past.
Add datetime/timezone import; update test assertion to match.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* perf(backend): convert delete_session os.remove() to async (#3812)

* perf(backend): convert delete_session os.remove() to async (#3796)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix: also await aiofiles.os.path.exists() in delete_session (review fix)

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(memory): wire WorkingMemory/EssentialStory/AgentDiary into UnifiedMemoryManager (#3822)

* feat(memory): wire WorkingMemoryService, EssentialStoryGenerator, AgentDiaryService into UnifiedMemoryManager (#3809)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(memory): wire WorkingMemoryService, EssentialStoryGenerator, AgentDiaryService into UnifiedMemoryManager (#3809)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(memory): make StandardizedAgent.memory_manager lazy-init

Avoid allocating UnifiedMemoryManager (SQLite + LRU cache) at every
agent construction — create it only on first property access.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(llm): consolidate llm_providers/ with llm_interface_pkg adapters (#3185)

- Add OTel tracing (Issue #697) and circuit breaker to
  llm_providers/openai_provider.py — previously dropped vs the older
  llm_interface_pkg/providers/openai_provider.py implementation
- Document that llm_providers/ollama_provider.py delegates
  chat_completion to llm_interface_pkg/providers/ollama.py which
  carries OTel tracing and circuit breaker; add docstring clarification
- Rewrite llm_interface_pkg/adapters/anthropic_adapter.py to delegate
  execute() to llm_providers.AnthropicProvider — eliminates ~90 lines of
  duplicated message-split/API-call logic; preserves test_environment()
- Rewrite llm_interface_pkg/adapters/openai_adapter.py to delegate to
  llm_providers.OpenAIProvider; fixes broken provider.api_key attribute
  reference in test_environment()
- Rewrite llm_interface_pkg/adapters/ollama_adapter.py to delegate to
  llm_providers.OllamaProvider instead of llm_interface_pkg/providers/ollama

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(backend): delete legacy chat_history_manager.py monolith (#3838)

Zero production importers — all call sites use `from chat_history import ChatHistoryManager`.
Superseded by chat_history/ mixin package since arch refactor #926.

* chore(backend): use config.path.docs_path in mcp_manual_integration (#3802) (#3817)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(autoresearch): deduplicate _NOTIFY_KEY format string (#3795) (#3818)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(ansible): move _SECRET_TO_ANSIBLE_VAR to shared location (#3778) (#3819)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(backend): add TimeoutConfig to SSOT config (#3820)

* chore(backend): add TimeoutConfig to SSOT config (#3803)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(backend): remove inline issue comments from TimeoutConfig callers

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(backend): wire additional TimeoutConfig fields — sandbox, llm_request, connection_pool (#3803)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(backend): consolidate terminal API implementations, remove compat aliases (#3383) (#3833)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* arch(knowledge): replace Redis adjacency list with queryable property graph (#3230) (#3844)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* arch(orchestration): unified graph model for DAG executor and LangGraph (#3228) (#3836)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(hardening): tasks 3-6 + perf(memory): fix O(n) get_all_facts scan (#3009, #3808) (#3846)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore: post-merge polish — hoist diary fetch limit, document property_graph serialisation, explain FeatureConfig alias suffixes

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(config): add alias to AutoBotConfig.path to prevent PATH env var collision (#3851)

* fix(orchestration): guard _save_checkpoint so failures never fail a successful step (#3825) (#3852)

chore(mcp): extract _cached_fetch helper, remove duplicated cache-check pattern (#3826)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(memory-graph): REST endpoints for invalidate_entity and invalidate_relation (#3810) (#3854)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* bug(memory): validate compression_threshold <= context_window_tokens at load (#3811) (#3855)

bug(a2a): evict terminal tasks after TTL to prevent TaskManager OOM (#3823)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(hardening): Tool SDK Registry + ToolRegistry SDK dispatch + PermissionEnforcementExtension (#3009, tasks 7-9) (#3853)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(hardening): WebSocket auth, shared WorkflowMemory, register PermissionExtension in lifespan (#3009, tasks 10-12) (#3857)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(llm): migrate chat completion calls to vLLM optimised API (#3389) (#3835)

- Fix bug in chat_completion_optimized(): was passing LLMRequest object to
  chat_completion(messages: list) — now calls _execute_with_fallback() +
  _finalize_response() directly so the pre-built provider="vllm" request
  is honoured
- Add automatic _calculate_cache_hit_rate() call in chat_completion() when
  the resolved provider is vllm, surfacing prefix-cache hit rates in
  response.metadata["cache_hit_rate"] for all vLLM-routed calls
- Expand AGENT_TIER_MAP with 7 task-agent types (summarization, translation,
  sentiment-analysis, code-generation, audio-processing, image-analysis,
  data-analysis) so get_base_prompt_for_agent() resolves them to Tier 3 instead
  of emitting a warning and falling back silently
- Migrate process_query() in all 7 task agents to chat_completion_optimized(),
  passing session_id / user_name / user_role from the request context; callers
  that carry conversation history (chat_agent) or custom multi-turn prompts
  (knowledge/RAG agents) are intentionally left on chat_completion() to
  preserve context fidelity

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* bug(backend): fix ChatHistoryManager create_task race with explicit initialize() (#3797) (#3815)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(agent-loop): content-aware repetitive tool-call detection (#3255) (#3832)

* feat(agent-loop): content-aware repetitive tool-call detection (#3255)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(agent-loop): correct off-by-one in repetition threshold check

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(config): config management enhancements — startup validation and sync API (#3398) (#3834)

* feat(config): config management enhancements — startup validation and sync API (#3398)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(config): replace blocking fdopen/json.dump with aiofiles in sync_config

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(ansible): add notify+flush_handlers to backend role — restart workers on autobot_shared reinstall (#3856)

* fix(chat): use ModelConfig.CHAT_NUM_CTX — was referencing wrong class ModelConstants

* refactor(chat): tombstone legacy conversation.py, remove dead ConversationManager from orchestrator (#3831) (#3865)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(voice): wire VoiceInterface into app.state, guard endpoints with 503 (#3848) (#3867)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(http): consolidate HTTP clients — TracedHttpClient wraps HTTPClientManager, extract sign_request (#3827) (#3870)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(autoresearch): wire real scorers + benchmark, add agent registration path (#3208) (#3876)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(knowledge): consolidate vector search under VectorSearchEngine (#3828) (#3879)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(a2a): guard AUTOBOT_A2A_CARD_TTL int() cast against invalid env var (#3824) (#3869)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(security): apply validate_path() to fix path-injection CodeQL alerts (#3164) (#3875)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(config): refactor sync_config to ≤65 lines per function (#3863) (#3888)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(backend): delete confirmed dead root-level modules (#3847) (#3887)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(types): rename 3 competing TaskComplexity enums to distinct names (#3878)

* chore(types): rename 3 competing TaskComplexity enums to distinct names (#3845)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(types): update test imports after TaskComplexity rename (#3845)

Replace TaskComplexity with ModelCapabilityTier in
utils/model_optimizer_refactoring_test.py to match the rename in
utils/model_optimization/types.py — fixes ImportError at pytest collection.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(backend): migrate singletons to AsyncInitializable lazy-init (#3885)

* refactor(backend): migrate singletons to AsyncInitializable lazy-init (#3390)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend): await async register_all_caches and get_cache_coordinator (#3390)

Three call sites in api/system.py and lifespan.py were missing await
after the functions were made async in the AsyncInitializable migration.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(llm): add lock to claude adapter singleton + reset _initialized on shutdown (#3390)

- Add _claude_adapter_lock to guard get_autobot_claude_adapter singleton
  creation against concurrent callers
- Reset self._initialized = False in shutdown() so ensure_initialized()
  re-runs after shutdown instead of fast-pathing into a None manager

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(agents): stream chain-of-thought events to frontend in real time (#3889)

* feat(agents): stream chain-of-thought events to frontend in real time (#3232)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(backend,frontend): CoT event session_id forwarding and activeSteps guard (#3232)

- mcp_dispatch: add session_id param to dispatch() and forward to emit calls
- cot_events: add done_callback to fire-and-forget create_task for exception logging
- useReasoningTrace: use Math.max(0, ...) to prevent activeSteps from going negative

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(agents): coroutine leak, missing step_complete on deny, frozen session ID (#3232)

- cot_events.py: move publish() call inline into loop.create_task() so
  the coroutine is only created after confirming a loop exists — fixes
  RuntimeWarning: coroutine was never awaited
- graph.py: add emit_step_complete() before early return on denied-
  approval path in execute_tools — fixes frontend activeSteps stuck at 1
- useReasoningTrace.ts: accept MaybeRef<string|null|undefined> instead
  of plain string so session ID is read reactively at handler call time

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(knowledge): resolve async Redis client coroutine ping error (#3872) (#3890)

get_redis_client(async_client=True) returns a coroutine because it delegates
to the async RedisConnectionManager.get_async_client() method. All 11 call
sites that assigned the result to a variable without await were storing a
coroutine object instead of a live Redis client, causing AttributeError on
any subsequent Redis operation (e.g. .ping(), .get(), .set()).

Fixed by adding await at each call site:
- knowledge/base.py (primary issue — .ping() failure on init)
- autobot_memory_graph/property_graph.py (initialize())
- initialization/lifespan.py (_wire_npu_task_queue())
- knowledge/search_components/reranking.py (_fetch_staleness_scores())
- services/autoresearch/osint_engine.py (_get_redis())
- services/autoresearch/prompt_optimizer.py (_get_redis())
- services/autoresearch/routes.py (4 route handlers)
- services/autoresearch/scorers.py (_get_redis())
- services/documentation_watcher.py (_on_file_changed())
- services/workflow_sharing_service.py (4 methods)
- services/workflow_versioning.py (4 methods)

* fix(ansible): deploy permission_rules.yaml to infrastructure config path (#3873) (#3891)

- Add permission_rules.yaml to backend role files/
- Ensure /opt/autobot/config exists before deploying the file
- Deploy permission_rules.yaml to /opt/autobot/config/ with notify: restart backend
- Remove the clean task that wiped /opt/autobot/config; that directory is
  required by permission_matcher.py at runtime (three levels up from services/)
- Update clean.yml comment to reflect that /opt/autobot/config is now live

* docs(specs): add language switcher design spec

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* docs(plans): add language switcher implementation plan

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(agent_loop): robust tool-call hash + repetition-halt guard (#3868 #3874 #3877)

Three related fixes to _compute_tool_call_hash and the halt mechanism:

- Issue #3868: wrap json.dumps() in try/except TypeError; use default=str as
  primary guard so non-serializable objects (custom classes, dataclasses) never
  raise — repr() fallback is only hit if default=str itself somehow fails.
- Issue #3874: non-dict args are now wrapped as {"__type__": ..., "__repr__": ...}
  instead of coerced to {}; distinct values (None, "", 42, "foo") now hash
  differently rather than all collapsing to the same empty-dict bucket.
- Issue #3877: add _halted_on_repetition flag (reset in _init_task_context);
  set it in _execute_tools when repetition fires; _should_continue() returns
  False immediately when flag is set, giving a belt-and-suspenders exit that
  works even if _should_iterate() misclassifies the halt error result.

Adds test_loop_repetition.py with 14 focused tests covering all three issues.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(autoresearch): rename test_issue_3208.py to benchmark_test.py (#3883) (#3893)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(config): use explicit None checks in startup validation (#3880) (#3897)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* chore(backend): delete dead type_definitions/ package (#3839) (#3892)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(agent_loop): don't add halted tool to executed list; surface error to LLM (#3859 #3862)

When the repetition-halt guard fires in _execute_tools(), the selected tool
never actually ran — yet _execute_iteration_phases was unconditionally adding
it to result.tools_executed and then falling through to _should_iterate()
which might not correctly classify the error.

Fix:
- Check _halted_on_repetition immediately after _execute_tools() returns.
- On halt: set result.tools_executed = [] (tool never ran, #3859) and
  result.tool_results = tool_results (error is visible to the LLM, #3862)
  then return early with should_continue=False.
- Normal (non-halt) execution path is unchanged.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.…