Project Organization Rules
Project Memory System
Project metadata is stored in the knowledge/projects/ directory as Agent "memory".
Memory Loading
When executing archival tasks, must first read:
knowledge/projects/_index.md
This index file contains:
- Active project list
- Project keyword mapping (for automatic file attribution)
- Project-related directory structure
Project-Based File Structure
knowledge/
├── projects/ # Project metadata (Agent Memory)
│ ├── _index.md # Project index (must read)
│ ├── {project-a}.md # Project A details
│ └── {project-b}.md # Project B details
├── programming/
│ ├── {project-a}/ # Project A - code related
│ ├── {project-b}/ # Project B - code related
│ └── commands/ # Common commands (no project affiliation)
├── biology/
│ ├── {project-a}/ # Project A - biology related
│ └── {project-b}/ # Project B - biology related
├── planning/
│ └── (organized by date/project TODO)
├── archive/ # Immutable inbox archive (by month)
│ └── YYYY-MM/
├── _memory/ # Agent memory (knowledge entries)
│ ├── _index.md # Knowledge index (must read for retrieval)
│ └── entries/
└── digests/ # Weekly review summaries
Rule Loading Guidance (On-Demand Loading)
Mandatory Rule Loading
When executing the following operations, must proactively use Read tool to load corresponding rule files:
| Operation Type | Must Load Rules |
|---|---|
| Archive inbox / organize files | knowledge/projects/_index.md (project index) |
| Create/edit planning files | notes-categorizer skill (planning section) |
| Create/edit programming files | notes-categorizer skill (programming section) |
| Create/edit biology files | notes-categorizer skill (biology section) |
| Handle sensitive information | private-info skill |
| Update changelog | changelog skill |
| Inbox processing (周回顾) | knowledge/_memory/_index.md + note-organizer skill (Inbox Processing section) |
| Search historical knowledge | knowledge/_memory/_index.md |
Why Proactive Loading?
Cursor's rule loading mechanism:
alwaysApply: true→ Always loadedalwaysApply: false+globs→ Only loaded when matching files are opened
Therefore, when processing inbox files, target directory rules won't auto-load; Agent must proactively read them.
Inbox Archival Workflow
- Load project index:
Read: knowledge/projects/_index.md - Analyze file content: Identify topics and keywords
- Match project: Determine project affiliation based on keyword mapping
- Load target rules: Read corresponding skill based on target directory
- Determine target path:
{category}/{project}/or{category}/ - Execute archival: Move file, update frontmatter, log to changelog
- Update project index: Update
_index.mdif new project created
Subdirectory Organization
To prevent directory clutter:
- Threshold: Create subdirectory when same-category files exceed 5
- Naming: Lowercase English, named by project/topic
- Resource files:
.assetsfolder in same directory as corresponding.mdfile
