hive-create

What it does

Provides an executable workflow to construct a goal-driven agent, including choosing a build path (from scratch or from a template), initializing a builder environment, registering an MCP server, listing available tools, and preparing a package structure for nodes. It then guides defining the agent goal with user collaboration and a rapid discovery pattern, culminating in capability assessment, gap analysis, and a recommendation framework. It is intended for use when actively building an agent and requires the "claude-code" tool.

How it works

• Step 0: Determine build path. If user hasn’t chosen, it asks to choose between:
  • From scratch: proceed to Step 1A
  • From a template: proceed to Step 1B
• STEP 1A (From Scratch):
  • Check for existing sessions with mcp__agent-builder__list_sessions()
  • If none, create a session via mcp__agent-builder__create_session(name="AGENT_NAME")
  • Register hive-tools MCP server with mcp__agent-builder__add_mcp_server(...) using stdio and a python server
  • Discover tools with mcp__agent-builder__list_mcp_tools()
  • Create package directory exports/AGENT_NAME/nodes
• STEP 1B (From Template):
  • List available templates by reading examples/templates/*/agent.json
  • Present templates to user and collect choice and agent name
  • Copy template to exports/NEW_AGENT_NAME
  • Create session, register MCP server, list tools, and load the template agent.json via mcp__agent-builder__import_from_export(agent_json_path="exports/NEW_AGENT_NAME/agent.json")
• STEP 2: Define Goal with user
  • If from template, present existing goal for approval; otherwise, prompt user to describe desired agent type and collaboratively shape the goal
  • Follow the discovery cadence (2a) to gather domain model, present a concrete proposed model, and ask 1-2 critical questions
• 2a to 2e: Capability assessment, gap analysis, and recommendation
  • Provide structured sections for good, bad, and ugly aspects relative to the framework’s capabilities, then identify gaps and offer a clear recommendation and explicit next step

When to use it

Use when actively building an agent and you want a guided, collaborative workflow that starts from build-path selection, environment setup, and ends with a goal definition and feasibility assessment.

What it can touch

The skill references interactions with the following tools and commands:

• "mcp__agent-builder__list_sessions()"
• "mcp__agent-builder__load_session_by_id(session_id="...")"
• "mcp__agent-builder__create_session(name="AGENT_NAME")"
• "mcp__agent-builder__add_mcp_server(...)"
• "mcp__agent-builder__list_mcp_tools()"
• Directory operations: "mkdir -p exports/AGENT_NAME/nodes"
• Template handling: reading "examples/templates/TEMPLATE_DIR/agent.json" and commands like "cp -r" to exports

Caveats

• License on the skill itself is Apache-2.0; behavior assumes access to the hive-tools MCP server and the claude-code tool.
• The workflow is executable and should be run as-is; it explicitly instructs not to explore code before starting and to proceed through the steps without displaying intermediate setup results.
• The success of 2a–2e depends on user collaboration and accurate interpretation of goals; the review emphasizes cautious, honest capability assessment and explicit next steps.