jamesgray-ai/registering-building-blocks

Registering AI Building Blocks

After creating, packaging, or updating any AI building block — Skill, Agent, Prompt, or Context MD — register or update it in the AI Building Blocks database.

Asset Type Resolution

Determine the asset type from the user's request before extracting metadata.

| Asset Type | Source Location | Identifier File | Name Source | |-----------|----------------|-----------------|-------------| | Skill | ~/.claude/skills/{name}/SKILL.md | SKILL.md frontmatter name field | Frontmatter name | | Agent | .claude/agents/{name}.md or ~/.claude/agents/{name}.md | Agent .md file frontmatter or filename | Filename (without .md) or frontmatter | | Prompt | User-specified file or inline text | N/A | User-provided name | | Context MD | CLAUDE.md or user-specified file | N/A | User-provided name or filename |

Resolution rules:

1. Keywords in the user request — "register this skill", "add this agent", "track this prompt", "register this context MD"
2. File path patterns — SKILL.md → Skill; .claude/agents/*.md → Agent; CLAUDE.md → Context MD
3. If ambiguous — Ask the user which asset type to use

Database Reference

• Data Source ID: 2d5edcfd-b924-80cf-a0a0-000ba0164e40
• Data Source URL: collection://2d5edcfd-b924-80cf-a0a0-000ba0164e40

Schema

| Property | Type | Purpose | |----------|------|---------| | Name | title | Building block name (from frontmatter, filename, or user input) | | Description | text | Description (from frontmatter or user input — includes trigger phrases for skills) | | Asset Type | select | "Skill", "Agent", "Prompt", or "Context MD" | | Platform | select | Always "Claude" for Claude building blocks | | Quick Start Prompt | text | Copy-paste prompt that demonstrates the building block | | GitHub | url | Repository URL if applicable |

Process

For each building block being registered:

Step 1: Extract Metadata

Based on the resolved asset type, read metadata from the appropriate source:

• Skill — Read SKILL.md frontmatter for name and description
• Agent — Read the agent .md file for name (filename or frontmatter) and description (opening paragraph or frontmatter)
• Prompt — Get name and description from the user or from the prompt file's content
• Context MD — Get name and description from the user; optionally summarize the file's purpose

Step 2: Generate Quick Start Prompt

Create a single, copy-paste-ready prompt that demonstrates the building block's primary use case.

Guidelines for Quick Start Prompts:

• One sentence that triggers the building block's main workflow
• Specific enough to be immediately useful
• Generic enough to work across different contexts
• Should produce a complete result, not just start a conversation

Examples by asset type:

| Type | Name | Quick Start Prompt | |------|------|--------------------| | Skill | reviewing-student-goals | "Navigate to my Maven course Reflect: Goals page and extract all student learning goals. Update each student's record in Notion and give me a cohort theme analysis." | | Skill | writing-linkedin-posts | "Write a LinkedIn post about [topic] using my brand voice." | | Agent | playbook-question-answerer | "Answer the question 'What are the six AI building blocks?' using the Hands-on AI site content." | | Agent | hbr-editor | "Review this article for HBR publication quality and give me prescriptive feedback." | | Prompt | meeting-prep-quick | "Prepare me for my meeting with [name] at [company] tomorrow." |

Context MDs typically don't need a Quick Start Prompt — leave blank or ask the user.

If unsure: Ask James for the Quick Start Prompt rather than guessing.

Step 3: Search for Existing Entry

Search the data source for the exact building block name:

{
  "query": "<building-block-name>",
  "data_source_url": "collection://2d5edcfd-b924-80cf-a0a0-000ba0164e40"
}

Critical: Check search results carefully. A match exists only if a result has the exact same title as the building block name. Partial matches or similar names are NOT duplicates.

Step 4: Create or Update

If exact match found → Update the existing page:

{
  "data": {
    "page_id": "<page-id-from-search>",
    "command": "update_properties",
    "properties": {
      "Description": "<description>",
      "Quick Start Prompt": "<generated-quick-start-prompt>"
    }
  }
}

If no exact match → Create new entry:

{
  "parent": {"data_source_id": "2d5edcfd-b924-80cf-a0a0-000ba0164e40"},
  "pages": [{
    "properties": {
      "Name": "<building-block-name>",
      "Description": "<description>",
      "Asset Type": "<Skill|Agent|Prompt|Context MD>",
      "Platform": "Claude",
      "Quick Start Prompt": "<generated-quick-start-prompt>"
    }
  }]
}

Batch Registration

When registering multiple building blocks (which can mix asset types):

1. Search for each building block individually before any creates/updates
2. Build two lists: items to update (with page IDs) and items to create
3. For each item, generate or request a Quick Start Prompt
4. Perform updates first, then batch create new entries
5. Report results: X created, Y updated (broken down by asset type)

Notes

• Always confirm registration with the user after modifying Notion
• If the building block has a GitHub repository, include the URL in the GitHub property
• Never create entries without first checking for duplicates
• If you can't determine an appropriate Quick Start Prompt, ask James for one