majiayu000 /auto-summary
Batch populate the summary frontmatter field on notes where it is null or missing. The summary field is the highest-value AI triage field — it enables fast note discovery without reading full content.
Usage
/auto-summary # Process all notes with null summary
/auto-summary <path> # Process specific file or folder
/auto-summary --dry-run # Preview summaries without applying
/auto-summary --type Concept # Only process specific note type
/auto-summary --limit 50 # Process at most 50 notes
Instructions
Phase 1: Find Notes Needing Summary
Use Grep to find notes with null or missing summary:
# Find notes with summary: null
Grep for "summary: null" in *.md files
# Also find notes with no summary field at all
# (These should have had summary added by template cleanup)
Filter results:
- Exclude
Templates/,.obsidian/,.claude/,Archive/,Attachments/ - Exclude Daily notes (journals don't need summaries)
- If
--typespecified, filter by frontmatter type - Sort by type for batch efficiency
Phase 2: Generate Summaries (Parallel)
Launch parallel Haiku sub-agents, batching 20 notes per agent:
For each note, the agent should:
- Read the full note (frontmatter + body)
- Generate a one-line summary following these rules:
Summary Writing Rules
- Length: 10-25 words. One sentence. No period at the end
- Voice: Active, descriptive. State what the note IS or DOES
- Content: Capture the core purpose, not details
- Avoid: Starting with "This note...", "A document about...", "Summary of..."
- Include: Key entities, technologies, or decisions where relevant
Summary Patterns by Type
| Type | Pattern | Example |
|---|---|---|
| Concept | What X is | Continuous Airworthiness Management Organisation responsible for aircraft safety compliance |
| Pattern | How to do X | Event-driven architecture pattern using Kafka for real-time system integration |
| Meeting | What was discussed/decided | Alpha sprint review covering data migration progress and API blockers |
| ADR | What was decided and why | Selected AWS Bedrock over Azure OpenAI for AlertHub safety processing |
| Project | What the project delivers | SAP to DataPlatform data integration enabling unified engineering analytics |
| System | What the system does | MRO Vendor MRO platform managing aircraft maintenance scheduling |
| Person | Role and context | Solutions Architect in Engineering IT, Alpha project lead |
| Task | What needs to be done | Implement Kafka consumer for Alpha work order events |
| Incubator | What idea is being explored | Exploring voice-activated Claude Code workflows for hands-free note capture |
| Research | What question was investigated | Analysis of vault structure identifying efficiency improvements for human and AI workflows |
| Reference | What the resource covers/teaches | AWS documentation on Bedrock guardrails for AI model safety |
| What the email communicates | Proposal to Beta programme board for Claude Code adoption across architecture team |
- Return — list of (filepath, summary) tuples
Phase 3: Apply Summaries
For each note with a generated summary:
- Read current frontmatter
- If
summary: null— replace with generated summary (quoted string) - If no summary field — add
summary:field after tags - Write updated file using Edit tool
Format:
summary: "Continuous Airworthiness Management Organisation responsible for aircraft safety compliance"
Always quote the summary value since it may contain special YAML characters (colons, brackets).
Phase 4: Report
## Auto-Summary Results
**Notes processed:** {{count}}
**Summaries added:** {{added_count}}
**Notes skipped:** {{skipped_count}} (already has summary or insufficient content)
### By Type
| Type | Summarised | Avg Length |
|------|-----------|------------|
| Concept | 45 | 15 words |
| Meeting | 80 | 18 words |
| ADR | 30 | 20 words |
### Sample Summaries
| Note | Summary |
|------|---------|
| {{note}} | {{summary}} |
| {{note}} | {{summary}} |
Safety
- Always use
--dry-runfirst for vault-wide operations - Never overwrite existing non-null summaries
- Commit to git before running
- If note body is too short (<50 words), skip rather than guess
- Summary is additive only — never removes existing summaries
Quality Checks
After running, verify quality by:
- Spot-check 10-15 summaries across different types
- Ensure summaries are accurate (not hallucinated)
- Check length (10-25 words target)
- Verify no YAML quoting issues
Related Skills
/summarize— Detailed single-note summarisation/quality-report— Includes summary coverage metrics/auto-tag— Companion batch field population skill