oGMemory/extraction · openGauss/oGMemory

Ccoconutnutxfix: discard pre-refetch candidates and fix RelationEdge field name

文件	最后提交记录	最后更新时间
prompts	fix: resolve rebase conflicts and integrate InternalToolUsageTracker into ReactLoop base class - Move _record_round/_record_tool_call from ExtractionReActLoop override to ReactLoop base class default implementation (tracker injected via __init__) - Fix indentation errors from conflict resolution in prefetch.py, api.py, extraction.yaml - Rename MemoryPrefetcher.prefetch to _prefetch (private), update all tests - Make ExtractionReActLoop.prompt_manager optional with fallback _build_system_prompt for no-template scenarios - Merge upstream extraction.yaml prompt with detail preservation rules Co-Authored-By: Claude GLM-5.1 <noreply@anthropic.com>	1 个月前
schemas	feat: define schema-driven extraction prompts	1 个月前
__init__.py	refactor: cleanup unused exports and obsolete verification scripts Remove unused validation re-exports from core/__init__.py, unused EXTRACTION_TOOLS/TOOL_TO_CATEGORY from extraction/__init__.py, unused collect_tool_stats from tool_collector.py, and obsolete scripts/verify_agfs_*.py files. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	3 个月前
chunking.py	feat: L0 structured summary generation + extraction prompt optimization Add compact-style L0 session summaries with 6 dimensions (events, facts, preferences, plans, emotions, key_details). Implement two-round retrieval: structured nodes for retrievedEvidence + session_summary for sessionContext. Strengthen extraction prompts for temporal precision and detail richness. Benchmark: run69 = 90.1% (385 questions, 3 samples), +8.5% over run68. Cat4 forgetting: 78.6% → 91.0%, Cat1 single: 84.4% → 93.2%, Cat2 cross: 78.4% → 86.7%. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2 个月前
extraction_react_loop.py	fix: discard pre-refetch candidates and fix RelationEdge field name - Mark extract_* tool records as superseded when safety refetch triggers, so pre-refetch candidates (produced without reading existing content) are excluded from final result collection in _to_extraction_react_result. - Fix AttributeError in _execute_get_relations: RelationEdge uses from_uri/to_uri, not target_uri. Lazy mode get_relations now returns correct field names instead of crashing. Co-Authored-By: Claude GLM-5.1[1m] <noreply@anthropic.com>	1 个月前
prefetch.py	fix: resolve rebase conflicts and integrate InternalToolUsageTracker into ReactLoop base class - Move _record_round/_record_tool_call from ExtractionReActLoop override to ReactLoop base class default implementation (tracker injected via __init__) - Fix indentation errors from conflict resolution in prefetch.py, api.py, extraction.yaml - Rename MemoryPrefetcher.prefetch to _prefetch (private), update all tests - Make ExtractionReActLoop.prompt_manager optional with fallback _build_system_prompt for no-template scenarios - Merge upstream extraction.yaml prompt with detail preservation rules Co-Authored-By: Claude GLM-5.1 <noreply@anthropic.com>	1 个月前
summary_generator.py	feat: L0 structured summary generation + extraction prompt optimization Add compact-style L0 session summaries with 6 dimensions (events, facts, preferences, plans, emotions, key_details). Implement two-round retrieval: structured nodes for retrievedEvidence + session_summary for sessionContext. Strengthen extraction prompts for temporal precision and detail richness. Benchmark: run69 = 90.1% (385 questions, 3 samples), +8.5% over run68. Cat4 forgetting: 78.6% → 91.0%, Cat1 single: 84.4% → 93.2%, Cat2 cross: 78.4% → 86.7%. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2 个月前
tool_builder.py	feat: formalize schema-driven extraction	1 个月前
tool_collector.py	feat: unify session usage tracking	1 个月前
tools.py	refactor: unify Lazy mode prompt with YAML template system - Add identity_block and lazy_tools_instruction to extraction.yaml - Lazy mode now uses system_prompt + lazy_tools_instruction from template - Remove hardcoded _build_system_prompt() in ExtractionReActLoop - prompt_manager becomes required parameter (always provided by Extractor) - Add reflection guidance for post-exploration phase in lazy_tools_instruction Co-Authored-By: Claude GLM-5[1m] <noreply@anthropic.com>	1 个月前