## Major Features Added - **Step-file workflow architecture**: Transform monolithic workflows into granular step files for improved LLM adherence and consistency - **Multi-menu handler system**: New `handler-multi.xml` enables grouped menu items with fuzzy matching - **Workflow compliance checker**: Added automated compliance validation for all workflows - **Create/edit agent workflows**: New structured workflows for agent creation and editing ## Workflow Enhancements - **Create-workflow**: Expanded from 6 to 14 detailed steps covering tools, design, compliance - **Granular step execution**: Each workflow step now has dedicated files for focused execution - **New documentation**: Added CSV data standards, intent vs prescriptive spectrum, and common tools reference ## Complete Migration Status - **4 workflows fully migrated**: `create-agent`, `edit-agent`, `create-workflow`, and `edit-workflow` now use the new granular step-file architecture - **Legacy transformation**: `edit-workflow` includes built-in capability to transform legacy single-file workflows into the new improved granular format - **Future cleanup**: Legacy versions will be removed in a future commit after validation ## Schema Updates - **Multi-menu support**: Updated agent schema to support `triggers` array for grouped menu items - **Legacy compatibility**: Maintains backward compatibility with single `trigger` field - **Discussion enhancements**: Added conversational_knowledge recommendation for discussion agents ## File Structure Changes - Added: `create-agent/`, `edit-agent/`, `edit-workflow/`, `workflow-compliance-check/` workflows - Added: Documentation standards and CSV reference files - Refactored: `create-workflow/steps/` with detailed granular step files ## Handler Improvements - Enhanced `handler-exec.xml` with clearer execution instructions - Improved data passing context for executed files - Better error handling and user guidance This architectural change significantly improves workflow execution consistency across all LLM models by breaking complex processes into manageable, focused steps. The edit-workflow transformation tool ensures smooth migration of existing workflows to the new format.
5.2 KiB
name, description, workflow_path, thisStepFile, agentFile, nextStepFile, advancedElicitationTask, partyModeWorkflow
| name | description | workflow_path | thisStepFile | agentFile | nextStepFile | advancedElicitationTask | partyModeWorkflow |
|---|---|---|---|---|---|---|---|
| step-04-apply-changes | Apply approved changes to the agent | {project-root}/src/modules/bmb/workflows/edit-agent | {workflow_path}/steps/step-04-apply-changes.md | {{agent_path}} | {workflow_path}/steps/step-05-validate.md | {project-root}/.bmad/core/tasks/advanced-elicitation.xml | {project-root}/.bmad/core/workflows/party-mode/workflow.md |
Step 4: Apply Changes
STEP GOAL:
Apply all user-approved changes to the agent files directly using the Edit tool.
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 an agent editor who helps users improve their BMAD agents through precise modifications
- ✅ If you already have 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 agent architecture expertise, user brings their agent and goals, together we improve the agent
- ✅ Maintain collaborative guiding tone throughout
Step-Specific Rules:
- 🎯 Focus only on applying changes that were explicitly approved in step 3
- 🚫 FORBIDDEN to make any changes that were not approved by the user
- 💬 Approach: Apply changes one by one with confirmation after each
- 📋 Use Edit tool to make precise modifications to agent files
EXECUTION PROTOCOLS:
- 🎯 Apply only changes that were explicitly approved in step 3
- 💾 Show confirmation after each change is applied
- 📖 Edit files directly using Edit tool with precise modifications
- 🚫 FORBIDDEN to make unapproved changes or extra modifications
CONTEXT BOUNDARIES:
- Available context: Approved changes list from step 3, agent path from step 1
- Focus: Apply ONLY the approved changes, nothing more
- Limits: Do not make any modifications beyond what was explicitly approved
- Dependencies: Must have approved changes list from step 3
Sequence of Instructions (Do not deviate, skip, or optimize)
1. Load Agent File
Read the complete agent file to understand current state before making changes.
2. Apply First Approved Change
For each change approved in step 3, apply it systematically:
For YAML changes in main agent file:
- Use Edit tool to modify the agent YAML file at
{agentFile} - Make the exact approved modification
- Confirm the change was applied correctly
For sidecar file changes (Expert agents):
- Use Edit tool to modify the specific sidecar file
- Make the exact approved modification
- Confirm the change was applied correctly
3. Confirm Each Change Applied
After each change is applied: "Applied change: {{description}}
- Updated section matches approved change ✓
- File saved successfully ✓"
4. Continue Until All Changes Applied
Repeat step 2-3 for each approved change until complete:
- Apply change using Edit tool
- Confirm it matches what was approved
- Move to next approved change
5. Verify All Changes Complete
"Summary of changes applied:
- {{number}} changes applied successfully
- All modifications match user approvals from step 3
- Agent files updated and saved"
6. 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 completion status to context, 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 [all approved changes from step 3 have been applied to agent files], will you then load and read fully {nextStepFile} to execute and begin validation of applied changes.
🚨 SYSTEM SUCCESS/FAILURE METRICS
✅ SUCCESS:
- All approved changes from step 3 applied using Edit tool
- Each modification matches exactly what was approved by user
- Agent files updated and saved correctly
- Confirmation provided for each applied change
- Menu presented and user input handled correctly
❌ SYSTEM FAILURE:
- Making changes that were not approved in step 3
- Using tools other than Edit tool for file modifications
- Not confirming each change was applied correctly
- Making extra modifications beyond approved changes
- Skipping confirmation steps or verification
Master Rule: Skipping steps, optimizing sequences, or not following exact instructions is FORBIDDEN and constitutes SYSTEM FAILURE.