Files
brachnha-insight/_bmad/bmb/workflows/agent/steps-e/e-01-load-existing.md
Max e9e6fadb1d fix: ChatBubble crash and DeepSeek API compatibility
- Fix ChatBubble to handle non-string content with String() wrapper
- Fix API route to use generateText for non-streaming requests
- Add @ai-sdk/openai-compatible for non-OpenAI providers (DeepSeek, etc.)
- Use Chat Completions API instead of Responses API for compatible providers
- Update ChatBubble tests and fix component exports to kebab-case
- Remove stale PascalCase ChatBubble.tsx file
2026-01-26 16:55:05 +07:00

6.4 KiB

name, description, thisStepFile, workflowFile, nextStepFile, editPlan, agentMetadata, agentMenuPatterns, advancedElicitationTask, partyModeWorkflow
name description thisStepFile workflowFile nextStepFile editPlan agentMetadata agentMenuPatterns advancedElicitationTask partyModeWorkflow
e-01-load-existing Load and analyze existing agent for editing ./e-01-load-existing.md ../workflow.md ./e-02-discover-edits.md {bmb_creations_output_folder}/edit-plan-{agent-name}.md ../data/agent-metadata.md ../data/agent-menu-patterns.md {project-root}/_bmad/core/workflows/advanced-elicitation/workflow.xml {project-root}/_bmad/core/workflows/party-mode/workflow.md

Edit Step 1: Load Existing Agent

STEP GOAL:

Load the existing agent file, parse its structure, and create an edit plan tracking document.

MANDATORY EXECUTION RULES (READ FIRST):

Universal Rules:

  • 🛑 NEVER proceed without loading the complete agent file
  • 📖 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 an autonomous editor
  • YOU MUST ALWAYS SPEAK OUTPUT In your Agent communication style with the config {communication_language}

Role Reinforcement:

  • You are an agent analyst who helps users understand and modify existing agents
  • 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 agent architecture expertise, user brings their modification goals, together we achieve successful edits
  • Maintain collaborative analytical tone throughout

Step-Specific Rules:

  • 🎯 Focus only on loading and analyzing the existing agent
  • 🚫 FORBIDDEN to make any modifications in this step
  • 💬 Approach: Analytical and informative, present findings clearly
  • 📋 Ensure edit plan is created with complete agent snapshot

EXECUTION PROTOCOLS:

  • 🎯 Load the complete agent YAML file
  • 📊 Parse and analyze all agent components
  • 💾 Create edit plan tracking document
  • 🚫 FORBIDDEN to proceed without confirming file loaded successfully

CONTEXT BOUNDARIES:

  • Available context: User provided agent file path from workflow
  • Focus: Load and understand the existing agent structure
  • Limits: Analysis only, no modifications
  • Dependencies: Agent file must exist and be valid YAML

MANDATORY SEQUENCE

CRITICAL: Follow this sequence exactly. Do not skip, reorder, or improvise unless user explicitly requests a change.

1. Load Agent File

Load the agent file: Read the complete YAML from the agent file path provided by the user.

If file does not exist or is invalid: Inform the user and request a valid path: "The agent file could not be loaded. Please verify the path and try again.

Expected format: {path-to-agent}/{agent-name}.agent.yaml"

2. Parse Agent Structure

If the module property of the agent metadata is stand-alone, it is not a module agent. If the module property of the agent is a module code (like bmm, bmb, etc...) it is a module agent. If the property hasSidecar: true exists in the metadata, then it is an expert agent. Else it is a simple agent. If a module agent also hasSidecar: true - this means it is a modules expert agent, thus it can have sidecar.

Extract and categorize all agent components:

# Basic Metadata
- name: {agent-name}
- description: {agent-description}
- module: {stand-alone|bmm|cis|bmgd|custom}
- hasSidecar: {true|false}

# Persona
- persona: {full persona text}
- system-context: {if present}

# Commands/Menu
- commands: {full command structure}

# Critical Actions (if present)
- critical-actions: {list}

# Metadata
- metadata: {all metadata fields}

3. Display Agent Summary

Present a clear summary to the user:

## Agent Analysis: {agent-name}

**Type:** {simple|expert|module}  (derived from module + hasSidecar)
**Status:** ready-for-edit

### Current Structure:

**Persona:** {character count} characters
**Commands:** {count} commands defined
**Critical Actions:** {count} critical actions

### Editable Components:

- [ ] Persona (role, identity, communication_style, principles)
- [ ] Commands and menu structure
- [ ] Critical actions
- [ ] Metadata (name, description, version, tags)

4. Create Edit Plan Document

Initialize the edit plan tracking file:

---
mode: edit
originalAgent: '{agent-file-path}'
agentName: '{agent-name}'
agentType: '{simple|expert|module}'
editSessionDate: '{YYYY-MM-DD}'
stepsCompleted:
  - e-01-load-existing.md
---

# Edit Plan: {agent-name}

## Original Agent Snapshot

**File:** {agent-file-path}
**Type:** {simple|expert|module}
**Version:** {version}

### Current Persona

{full persona text or truncated if very long}

### Current Commands

{list all commands with names and descriptions}

### Current Metadata

{all metadata fields}

---

## Edits Planned

*This section will be populated in subsequent steps*

---

## Edits Applied

*This section will track completed edits*

Write to {editPlan}.

5. Present MENU OPTIONS

Display: "Is this the correct agent to edit? [C] Yes, Continue to Discovery"

Menu Handling Logic:

  • IF C: Save content to {editPlan}, 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 [agent file loaded, analyzed, and edit plan created], will you then load and read fully {nextStepFile} to execute and begin edit discovery.


🚨 SYSTEM SUCCESS/FAILURE METRICS

SUCCESS:

  • Agent file loaded successfully
  • YAML structure parsed correctly
  • Edit plan document created with agent snapshot
  • User has clear understanding of current agent structure
  • Menu presented and user input handled correctly

SYSTEM FAILURE:

  • Failed to load entire exist agent file (and potential sidecar content)
  • Invalid YAML format that prevents parsing
  • Edit plan not created
  • Proceeding without user confirmation of loaded agent

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