mirror of
https://github.com/agentscope-ai/ReMe.git
synced 2026-10-11 03:40:03 +00:00
ffb4d08c4f
* feat(file_io): add daily_write step for creating daily notes with conversation metadata - Add DailyWriteStep class that delegates to write job for creating daily notes - Register daily_write job in default configuration with proper parameters - Include validation for name and session_id path components - Add test coverage for daily_write functionality including metadata handling - Preserve existing job execution method in application.py after repositioning - Update base_step.py to use positional-only parameter syntax for job methods - Import and expose DailyWriteStep in file_io module initialization - Override reserved metadata keys (name, description, session_id, source_conversation) with fixed values - Refresh daily index after successful write operation - Generate proper source conversation links in markdown format * feat(daily): refactor daily note system with enhanced metadata handling - Introduce validate_filename_component function and export it - Add _INDEX_HIDDEN_METADATA_KEYS to hide conversation metadata from index - Update scan_notes to exclude hidden metadata keys from index rendering - Modify auto_memory to use daily_write tool and manage session frontmatter - Implement session note lookup and renaming based on frontmatter name - Update daily_list to return flattened note metadata including session info - Change daily_write to dispatch write step instead of running job - Add test cases for updated daily note functionality and metadata handling - Update version from 0.4.0.2 to 0.4.0.3 * fix(evolve): correct metadata update in auto memory response - Fixed trailing comma issue in metadata dictionary update - Ensured proper formatting of response metadata structure - Maintained existing functionality while fixing syntax error * refactor(auto_resource): replace daily_create with dynamic note management - Remove DailyCreateStep and related exports from file_io module - Replace static daily note creation with dynamic resource-linked card system - Implement LLM-suggested naming with frontmatter-driven file management - Add source_resource linking for tracking original files - Introduce collision handling with hash-based suffixes - Update documentation to reflect new resource card workflow - Modify auto_resource prompts to use write/edit tools instead of daily_create - Adjust test fixture comments to match new agent behavior - Update framework diagrams and quick start examples accordingly * feat(app): add version info to app initialization and update auto-memory logic - Include version number in application startup logging - Remove tool result truncation logic from auto-memory step - Update auto-memory to exclude tool_result blocks from saved history - Add test case to verify tool results are filtered out from message saving - Update YAML prompts to clarify filename naming rules without dates - Modify configuration to support new dispatch steps format with persistence control * feat(auto_memory): add note modification tracking and optimize frontmatter updates - Add _note_bytes and _note_modified methods to track actual file changes - Optimize frontmatter updates by checking existing metadata before update - Add modified flag to response metadata indicating actual note changes - Update logging to include modified status in various operations - Add comprehensive tests for modified/unmodified detection scenarios - Enhance result hook logic to skip when no actual changes occur - Refactor metadata handling to properly track creation vs modification status
3.4 KiB
3.4 KiB
Auto Resource Beta
Auto Resource 是 ReMe 的资源解读入口，目前处于 Beta。资源文件先按日期进入 resource/，再被解读成 daily
资源卡片；卡片文件名由 LLM 生成的 frontmatter name 决定，并通过 source_resource 追溯原始文件。
关于 workspace 分层、resource/ 和 daily/ 的通用文件语义，见 Memory as File。对话进入 daily 的流程见
Auto Memory。
resource/YYYY-MM-DD/<resource_file>
├─ step 1: daily/YYYY-MM-DD/<generated_name>.md # 资源解读卡片
├─ step 2: source_resource 指回原始资源
└─ step 3: daily/YYYY-MM-DD.md # 当天索引再串起来
它记录什么
它不只是搬运文件内容，而是把资料里以后方便检索和理解的信息提炼出来：
- 核心内容：这份资料主要讲什么。
- 结构脉络：章节、表格、字段、数据组织方式。
- 关键细节：重要数字、名称、日期、结论。
- 背景用途：这份资料为什么存在，和当前工作有什么关系。
- 可行动项：任务、截止时间、后续跟进。
简单说，它负责把“文件存档”变成“资料可用”。
原始资料入口
Auto Resource 以 resource/ 作为原始资料入口。资源需要按日期放置，这个日期会决定它进入哪一天的 daily 记忆层。
示例目录：
workspace/
resource/
2026-06-20/
market-report.md
meeting-notes.csv
当前 Beta 版本更适合处理文本类资源，例如 md、txt、json、jsonl、csv、yaml、html。
资源卡片
每个资源文件会生成一张 daily 资源卡片。创建时先使用资源文件 stem 作为临时路径，Agent 写入后，系统会根据
frontmatter name 重命名文件：
resource/2026-06-20/market-report.md
↓
daily/2026-06-20/市场报告要点.md
资源卡片通过 frontmatter 关联原始文件：
source_resource: "[[resource/2026-06-20/market-report.md]]"
如果资源文件更新，Auto Resource 会通过 source_resource 找到对应卡片并更新；如果资源文件删除，对应的 daily note
也会被清理。旧版本按 stem 生成的 daily/YYYY-MM-DD/<resource_stem>.md 仍作为 fallback 兼容。
当天索引
资源卡片会进入和 Auto Memory 相同的 daily 记忆层。当天的 YYYY-MM-DD.md 会作为索引页，把这些资源卡片组织起来：
daily/
2026-06-20.md
2026-06-20/
市场报告要点.md
会议纪要整理.md
以后想回看这一天处理过哪些资料，先看 YYYY-MM-DD.md；想看某份资料沉淀了什么，再进入对应的资源卡片。
同时保留原始资料
解读后的 daily note 负责“好读”，原始资源负责“可信”。
Auto Resource 不会把原始文件挪走：它仍然留在 resource/YYYY-MM-DD/。这样，文本资料会进入 daily 记忆流，原始文件也始终保留在它来时的位置。
后续流向
Auto Resource 只生成 daily 层的资源解读。要把资源中的长期知识沉淀进 digest/，使用 Auto Dream；要检索原始资源、
daily 卡片和 digest 节点，使用 Memory Search。