agents_split_reference.md

agents.py split reference

• Backup: backups/agents.py.pre_split.20260313_141234.bak
• Compatibility surface: from agents import Agent remains unchanged.
• Bound methods keep the agents.py module globals so existing monkeypatch-based tests can keep targeting agents.

Extracted modules

• model_methods.py: _select_generation_mode, _select_response_model, _model_role_for_generation_mode, _model_retryable_error_markers, _is_retryable_model_error, _record_model_success, _record_model_failure, _model_candidate_chain, _chat_with_model_failover
• text_methods.py: _clean_think_tags, _strip_unwanted_json, _looks_like_tool_dump, _strip_search_meta_leakage, _strip_internal_prompt_leakage, _naturalize_search_output
• history_methods.py: _should_use_per_user_history, _get_history_list, _history_limits, _estimate_text_tokens_rough, _history_token_estimate, _resolve_history_compaction_thresholds, _state_ledger_for_user, _state_ledger_block, _history_for_prompt, _compact_history_if_needed, _push_history_for, _tool_loop_config, _tool_call_history, _run_tool_with_loop_guard
• search_memory_methods.py: _format_due_ts_local, _is_personal_memory_query, _should_inject_due_context, _mark_due_context_injected, _route_local_memory_intents, _should_websearch, _build_rag_block, _maintenance_tick, _persist_memory_serial, _memory_scope_for_prompt, _response_timeout_seconds, _vision_timeout_seconds, _token_budget, _extract_urls_from_results, _extract_urls_from_citation_map, _enforce_web_evidence_output, _is_volatile_results, _safe_build_image_spec, _is_chart_keyword_prompt, _numeric_guard, _is_finance_intent_hint, _looks_like_historical_price_followup, _looks_like_finance_followup, _render_direct_finance_answer, _render_direct_volatile_answer
• response_methods.py: generate_response, _set_last_attachments, get_last_attachments, generate_response_with_attachments, analyze_image