Security:
- H1 DNS rebinding TOCTOU: validate_endpoint returns (endpoint, resolved_ips);
create_pinned_session builds an aiohttp session with a custom resolver that
returns only the pre-validated IPs, closing the re-resolve gap.
- H2/M3 Shared session cookie pollution: integration no longer uses HA shared
clientsession; isolated session with DummyCookieJar prevents cross-integration
cookie leaks.
- Cloud metadata / link-local block added to allow_local_network mode to
prevent IMDS exfiltration on cloud VMs.
- L3 hard cap extended to history.async_get_history (not only service schema).
ML/LLM correctness:
- _is_openai_reasoning_model uses regex with gpt-5-chat* blacklist and handles
OpenRouter-style openai/ prefix. Future o5/gpt-6 auto-recognized.
- reasoning_effort=minimal for gpt-5 family, low for o-series.
- Gemini 2.5 Pro gets thinking_budget=128 (Pro rejects 0, flash accepts 0).
- Anthropic extracts first type=text block instead of hardcoded content[0].
- /no_think dedup uses word-boundary regex instead of substring.
- DeepSeek-reasoner detected: skips /no_think; preserves reasoning_content.
Reliability:
- Exception chaining added to Gemini-block re-raises.
- Top-level imports for json/re/hashlib (no more lazy stdlib imports).
- allow_local_network logged at INFO, not WARNING on every setup.
Code style:
- PEP 604 type hints across all .py files (dict/list/| None).
No breaking changes for users. Internal API: validate_endpoint return type
changed from str to tuple[str, list[str]].
Replace Creative Commons Attribution-NonCommercial-ShareAlike 4.0
with PolyForm Noncommercial License 1.0.0 across all files:
LICENSE, README badge/footer, and all Python module headers.
Phase A — Critical:
- Remove dual timeout stacking in coordinator._send_to_api
- Add Gemini-specific asyncio.timeout (sync SDK via to_thread)
- Store full text in history for context; cap per-field at 32KB on disk
- Fix instance lookup to match by normalized_name
- Add Bearer/sk-/x-api-key credential sanitization patterns
Phase B — Dead code removal:
- Merge async_ask_question/async_process_question into single method
- Remove dead is_anthropic flag from coordinator and __init__
- Remove unused DEFAULT_TIMEOUT and API_TIMEOUT constants
- Remove redundant _create_history_dir calls
- Consolidate async_check_api to use provider registry
Phase C — Config flow correctness:
- Truncate name before uniqueness check (prevent post-truncation collisions)
- Add async_set_unique_id + _abort_if_unique_id_configured
- Extract shared _build_parameter_schema for ConfigFlow/OptionsFlow dedup
Phase D — UX improvements:
- Optimize history write: serialize from memory, single file write
- Show last 5 history entries in sensor attributes (was 1)
- Return actual error type in ask_question service response
- Add dedicated api_key_required error for provider/endpoint changes
- Pass config_entry to DataUpdateCoordinator (HA 2024.8+)
Phase E — Cleanup:
- Extract _apply_structured_output for OpenAI/DeepSeek dedup
- Reduce ABSOLUTE_MAX_HISTORY_SIZE to 200 with Final annotation
- Remove dead translation keys (queued, invalid_characters)
- Migrate to _attr_has_entity_name = True
- Add from __future__ import annotations to all modules
- Remove redundant api_status sensor attribute
- Add missing translation keys (last_model, last_timestamp, etc.)
Review agent fixes:
- Add archive file cleanup (max 3 archives) to prevent disk exhaustion
- Per-entry storage cap (32KB per field) for history on disk
SMKRV