auto-tag - SKILL.md Agent Skill

name: auto-tag context: fork skill: auto-tag model: haiku description: Batch auto-tag untagged notes using content analysis tags: [activity/governance, domain/tooling, automation]

/auto-tag

Batch auto-tag notes that have empty or missing tags by analysing their content and type. Addresses the 61% empty tag rate in the vault.

Usage

/auto-tag                    # Process all untagged notes
/auto-tag <path>             # Process specific file or folder
/auto-tag --dry-run          # Preview changes without applying
/auto-tag --type Meeting     # Only process untagged notes of a specific type
/auto-tag --limit 50         # Process at most 50 notes

Instructions

Phase 1: Find Untagged Notes

Use Grep and Glob to find notes with empty tags:

# Find files with empty tags arrays
Grep for "tags: \[\]" in *.md files (exclude Templates/, .obsidian/, .claude/, Archive/)

# Also find files with no tags field at all
# Notes with tags containing only template variables

Filter results:

Exclude Templates/, .obsidian/, .claude/, Archive/, Attachments/
If --type specified, filter by frontmatter type field
If <path> specified, scope to that path
Sort by type for batch efficiency

Phase 2: Analyse and Tag (Parallel)

Launch parallel Haiku sub-agents, batching 20 notes per agent:

For each note, the agent should:

Read frontmatter — extract type, title, project, status
Read body content — first 500 words
Determine tags using these rules:

Tag Selection Rules

By note type (always apply):

Type	Auto-tags
ADR	`type/adr`, `activity/architecture`
Meeting	(infer from title/content)
Task	(infer project from `project:` field)
Project	`activity/delivery` + self-referencing `project/` tag
System	`type/system`
Incubator	`activity/research`
Research	`activity/research`
Concept	(infer from content)
Pattern	(infer from content)
Email	(infer from content)
Reference	(infer from referenceType and content)
Trip	(no tags needed — skip)
Daily	`daily`

By content analysis (keyword matching):

Content Keywords	Suggested Tag
AWS, Lambda, S3, EC2, Bedrock	`technology/aws`
SAP, BTP, S/4HANA, UI5	`technology/sap`
ERPSystem, MRO Vendor	`technology/erp`
Kafka, streaming, event	`technology/kafka`
Snowflake, data warehouse	`technology/snowflake`
security, IAM, encryption	`domain/security`
data, analytics, pipeline	`domain/data`
engineering, maintenance, MRO	`domain/engineering`
integration, API, middleware	`domain/integration`
cloud, infrastructure	`domain/cloud`
architecture, design, pattern	`activity/architecture`
research, investigation, POC	`activity/research`
governance, compliance, policy	`activity/governance`

By project: field (if present):

Project Link	Tag
`[[Project - Alpha]]`	`project/alpha`
`[[Project - Beta]]` or `[[Project - Beta]]`	`project/beta`
`[[Project - AlertHub]]`	`project/alerthub`
`[[Project - Gamma]]`	`project/gamma`
`[[Project - Delta]]`	`project/delta`
`[[Project - Epsilon]]`	`project/epsilon`
`[[Project - Cyber Uplift]]`	`project/cyber-uplift`

Tag count target: 2-5 tags per note. Prefer fewer, more accurate tags over many vague ones.

Format tags — all lowercase, hierarchical, no # prefix
Return — list of (filepath, suggested_tags) tuples

Phase 3: Apply Tags

For each note with suggested tags:

Read current frontmatter
If tags: [] — replace with suggested tags
If no tags field — add tags: with suggested tags
Write updated file using Edit tool

Format:

# Inline for 1-3 tags:
tags: [activity/research, domain/data]

# Multi-line for 4+ tags:
tags:
  - activity/architecture
  - technology/aws
  - domain/cloud
  - project/alerthub

Phase 4: Report

## Auto-Tag Results

**Notes processed:** {{count}}
**Notes tagged:** {{tagged_count}}
**Notes skipped:** {{skipped_count}} (already tagged or no suggestions)

### By Type
| Type | Tagged | Example Tags |
|------|--------|-------------|
| Meeting | 45 | project/*, domain/* |
| Concept | 30 | activity/*, domain/* |
| ADR | 20 | type/adr, technology/* |

### Sample Changes
| Note | Tags Added |
|------|-----------|
| {{note}} | {{tags}} |
| {{note}} | {{tags}} |

Safety

Always use --dry-run first for vault-wide operations
Never overwrite existing non-empty tags
Commit to git before running
Tags are additive only — never removes existing tags
Use the .claude/context/tag-taxonomy.md reference for valid hierarchies

Related Skills

/tag-management — Manual tag analysis and refactoring
/quality-report — Includes tag coverage metrics
/vault-maintenance — Includes tag health checks