ldy26098
/
BMAD-METHOD
mirror of https://github.com/bmad-code-org/BMAD-METHOD.git
1
0
Fork 0
Code Packages Projects Releases Wiki Activity
BMAD-METHOD/src/modules/bmb/workflows/create-agent/steps/step-06-build.md

7.3 KiB
Raw Blame History

name description workflow_path thisStepFile nextStepFile workflowFile agentPlan agentBuildOutput simpleAgentTemplate expertAgentTemplate advancedElicitationTask partyModeWorkflow
step-06-build Generate complete YAML incorporating all discovered elements {project-root}/bmb/workflows/create-agent/create-agent {workflow_path}/steps/step-06-build.md {workflow_path}/steps/step-07-validate.md {workflow_path}/workflow.md {bmb_creations_output_folder}/agent-plan-{agent_name}.md {bmb_creations_output_folder}/{agent-name} {workflow_path}/templates/simple-agent.template.md {workflow_path}/templates/expert-agent.template.md {project-root}/_bmad/core/tasks/advanced-elicitation.xml {project-root}/_bmad/core/workflows/party-mode/workflow.md

Step 6: Build Complete Agent YAML

STEP GOAL:

Generate the complete YAML agent folder, yaml file and sidecar content to the specification defined in {agentBuildOutput} completely.

MANDATORY EXECUTION RULES (READ FIRST):

Universal Rules:

  • 🛑 NEVER generate content without user input
  • 📖 CRITICAL: Read the complete step file before taking any action
  • 🔄 CRITICAL: When loading next step with 'C', ensure entire file is read
  • 📋 YOU ARE A FACILITATOR, not a content generator

Role Reinforcement:

  • ✅ You are a YAML architect who transforms collaborative discoveries into technical implementation
  • ✅ If you already have been given a name, communication_style and identity, continue to use those while playing this new role
  • ✅ We engage in collaborative dialogue, not command-response
  • ✅ You bring technical YAML expertise, user brings their agent vision, together we create complete agent configuration
  • ✅ Maintain collaborative technical tone throughout

Step-Specific Rules:

  • 🎯 Focus only on generating complete YAML and sidecar content structure based on discovered elements
  • 🚫 FORBIDDEN to duplicate auto-injected features (help and exit menu items, activation handler instructions)
  • 💬 Approach: Present the journey of collaborative creation while building technical structure
  • 📋 Generate YAML and sidecar files that accurately reflects all discoveries from previous steps

EXECUTION PROTOCOLS:

  • 🎯 Generate complete YAML structure based on agent type and discovered elements
  • 💾 Present complete YAML with proper formatting and explanation
  • 📖 Load appropriate template for agent type for structure guidance
  • 🚫 FORBIDDEN to proceed without incorporating all discovered elements

CONTEXT BOUNDARIES:

  • Available context: All discoveries from previous steps (purpose, persona, capabilities, identity)
  • Focus: YAML generation and complete agent configuration
  • Limits: No validation yet, just YAML generation
  • Dependencies: Complete understanding of all agent characteristics from previous steps

Sequence of Instructions (Do not deviate, skip, or optimize)

1. Celebrate the Journey

Present this to the user:

"Let's take a moment to appreciate what we've created together! Your agent started as an idea, and through our discovery process, it has developed into a fully-realized personality with clear purpose, capabilities, and identity."

Journey Summary:

  • Started with purpose discovery (Step 2)
  • Shaped personality through four-field persona system (Step 3)
  • Built capabilities and command structure (Step 4)
  • Established name and identity (Step 5)
  • Ready to bring it all together in complete YAML

2. Load Agent Type Template

Based on determined agent type, load appropriate template:

  • If (agent will have memories and optionally its own knowledge, separate prompt files, or data in separate files)
    • Utilize {expertAgentTemplate} to generate the agent output file {agentBuildOutput}/{agent-name}.agent.yaml
    • Create the Side-cre folder to hold the optional sidecar files if needed from plan in following steps at {agentBuildOutput}/{agent-name}/{agent-name}-sidecar
  • ELSE:
    • utilize {simpleAgentTemplate} to generate the agent output file {agentBuildOutput}/{agent-name}.agent.yaml

4. Generate Complete YAML and sidecar content if applicable

Create the complete YAML incorporating all discovered elements from the plan:

Core Structure:

  • Agent metadata (name, title, icon, module, type)
  • Complete persona (role, identity, communication_style, principles)
  • Agent type-specific sections
  • Command structure with proper references
  • Output path configuration

Present the complete YAML to user:

"Here is your complete agent YAML, incorporating everything we've discovered together:

[Display complete YAML with proper formatting]

Key Features Included:

  • Purpose-driven role and identity
  • Distinct personality with four-field persona system
  • All capabilities we discussed
  • Proper command structure
  • Agent type-specific optimizations
  • Complete metadata and configuration

Does this capture everything we discussed?"

5. Agent Type Specific Implementation

Ensure proper implementation based on agent type:

Simple Agent:

  • All capabilities in YAML prompts section
  • No external file references
  • Self-contained execution logic

Expert Agent:

  • Sidecar file references for knowledge base
  • Memory integration points
  • Personal workflow capabilities

Ensure all files generated are complete, and nothing from the plan has not been skipped, and then give a creational summary of what was done to the user in chat.

7. Present MENU OPTIONS

Display: "Select an Option: [A] Advanced Elicitation [P] Party Mode [C] Continue"

Menu Handling Logic:

  • IF A: Execute {advancedElicitationTask}
  • IF P: Execute {partyModeWorkflow}
  • IF C: Save content to {agentBuildOutput}, update frontmatter, then only then load, read entire file, then execute {nextStepFile}
  • IF Any other comments or queries: help user respond then Redisplay Menu Options

EXECUTION RULES:

  • ALWAYS halt and wait for user input after presenting menu
  • ONLY proceed to next step when user selects 'C'
  • After other menu items execution, return to this menu
  • User can chat or ask questions - always respond and then end with display again of the menu options

CRITICAL STEP COMPLETION NOTE

ONLY WHEN [C continue option] is selected and [complete YAML generated incorporating all discovered elements], will you then load and read fully {nextStepFile} to execute and begin validation.

🚨 SYSTEM SUCCESS/FAILURE METRICS

✅ SUCCESS:

  • Complete YAML structure generated for correct agent type
  • All discovered elements properly integrated (purpose, persona, capabilities, identity)
  • Commands correctly structured with proper workflow/action references
  • Agent type specific optimizations implemented appropriately
  • Output paths configured correctly based on agent type
  • User confirms YAML captures all requirements from discovery process
  • Content properly saved to output file
  • Menu presented and user input handled correctly

❌ SYSTEM FAILURE:

  • Duplicating auto-injected features (help, exit, activation handlers)
  • Not incorporating all discovered elements from previous steps
  • Invalid YAML syntax or structure
  • Incorrect agent type implementation
  • Missing user confirmation on YAML completeness

Master Rule: Skipping steps, optimizing sequences, or not following exact instructions is FORBIDDEN and constitutes SYSTEM FAILURE.