Template Assistant Skill

Workflow

0. Gather Vault Context (NEW)

Before creating notes, understand what already exists:

# List existing areas
bun ${CLAUDE_PLUGIN_ROOT}/src/cli.ts list-areas --format json

# List existing projects (for task linking)
bun ${CLAUDE_PLUGIN_ROOT}/src/cli.ts list-projects --format json

# Show allowed tags from config
bun ${CLAUDE_PLUGIN_ROOT}/src/cli.ts list-tags --format json

CRITICAL: Classification vs Invention

When selecting areas/projects, you are CLASSIFYING content into existing categories, NOT inventing factual data:

• Areas/Projects = CLASSIFICATION (analytical task)

  • Analyze the content to determine which life domain or responsibility it belongs to
  • Areas are ongoing RESPONSIBILITIES or LIFE DOMAINS (Home, Work, Health, Finance, Learning, Family, etc.)
  • Projects are temporary initiatives with completion dates
  • Example classifications:
    • Garden shed construction → [[Home]] (ongoing home maintenance responsibility)
    • Fitness goals → [[Health]] (ongoing health management domain)
    • Work deadline → [[Work]] (professional responsibilities)

• Factual Data = INVENTION (requires user knowledge)

  • Dates, numbers, specific names must come from user
  • Use null when unknown, never guess

Via MCP tools:

• para_list_areas - Get existing areas
• para_list_projects - Get existing projects
• para_list_tags - Get allowed tags
• para_scan_tags - See tags actually in use

For slash commands: Use AskUserQuestion to present existing areas/projects as options with descriptions of what domain they represent. Include "Other" for new classifications when content doesn't fit existing categories.

1. Discover Template Structure

bun ${CLAUDE_PLUGIN_ROOT}/src/cli.ts template-fields project --format json

Returns required args, auto-filled fields, and body sections.

2. Gather User Context

Ask focused questions matching the template type:

Classification approach:

• Analyze note content to determine life domain or project context
• Present existing areas/projects as classification options
• Only suggest new areas/projects when content clearly doesn't fit existing categories
• Remember: Classification = analytical judgment, not data invention

3. Generate Section Content

Content-Heavy (project, resource, weekly-review, daily): Generate paragraphs, bullet lists, [[wikilinks]]

Metadata-Heavy (task, booking, checklist, capture): Focus on frontmatter, minimal body content

CRITICAL: When generating wikilinks for frontmatter args, do NOT include quotes:

• ✅ Correct: --arg "Area=[[Product]]"
• ❌ Wrong: --arg "Area=\"[[Product]]\""

This ensures Dataview queries work correctly.

4. Create Note with Content

bun ${CLAUDE_PLUGIN_ROOT}/src/cli.ts create --template project \
  --title "Launch Dark Mode" \
  --arg "Area=[[Product]]" \
  --arg "Target completion date (YYYY-MM-DD)=2025-03-31" \
  --content '{"Why This Matters": "Dark mode reduces eye strain...", "Success Criteria": "- [ ] Theme toggle works\n- [ ] Persists across sessions"}'

5. Validate Result

bun ${CLAUDE_PLUGIN_ROOT}/src/cli.ts frontmatter validate "Launch Dark Mode.md" --format json

Error Handling

References

Load these as needed based on the task:

• Template details: ./references/template-catalog.md — Full catalog of all 11 templates with required args and body sections
• Generation patterns: ./references/content-strategies.md — Template-specific content generation strategies (goal clarification, success criteria, risk identification)
• Examples: ./references/examples.md — Complete CLI examples for project, area, resource, task, capture

Vault-Aware Workflows

Automatic Mode (convert command)

The convert command uses classification-based prompting to intelligently populate area/project fields:

bun ${CLAUDE_PLUGIN_ROOT}/src/cli.ts convert note.md --template project
# Vault context: 5 areas, 12 projects, 20 tags
# LLM analyzes content and classifies into existing [[Health]] area
# Example: "fitness tracking" content → [[Health]] (not "Wellness" or "Fitness")

How classification works:

1. LLM receives existing areas/projects as classification options
2. Analyzes note content to determine life domain or project context
3. Selects best-matching existing category OR suggests new one if content doesn't fit
4. Areas represent ongoing RESPONSIBILITIES (Home, Work, Health) not temporary topics
5. Projects represent temporary INITIATIVES with completion dates

Classification examples from prompt:

• Garden shed → [[Home]] (ongoing home maintenance responsibility)
• Fitness goals → [[Health]] (ongoing health management domain)
• Work project → [[Work]] (professional responsibilities domain)

Interactive Mode (slash commands)

For manual note creation via slash commands, use classification-based questioning:

1. Fetch context via MCP tools
2. Analyze content to determine likely domain
3. Present options with domain descriptions via AskUserQuestion
4. Create note with classification

Example:

// 1. Fetch existing areas
const { areas } = await para_list_areas({ response_format: "json" });

// 2. Analyze content and ask user (with domain context)
const answer = await AskUserQuestion({
  question: "This content appears to be about fitness tracking. Which life domain should it belong to?",
  header: "Area",
  options: [
    { label: "Health", description: "Ongoing health & wellness management (recommended for fitness)" },
    { label: "Personal", description: "Personal development & self-improvement" },
    { label: "Home", description: "Home responsibilities & maintenance" }
  ],
  multiSelect: false
});

// 3. Create with classified area
await para_create({
  template: "project",
  title: "My Project",
  args: { "Area": `[[${answer}]]` }  // No quotes!
});

Classification guidance:

• Frame options with domain descriptions so user understands what each area represents
• Suggest most likely classification based on content analysis
• Only offer "Other" when content clearly doesn't fit any existing domain

Tag Selection Pattern

Tags are hard constrained - must come from config:

# Get allowed tags
const { tags } = await para_list_tags({ response_format: "json" });

# LLM must choose from: project, area, resource, task, daily, journal...
# NO new tags allowed