- 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
6.3 KiB
name, description, nextStepFile, editPlan, advancedElicitationTask, partyModeWorkflow
| name | description | nextStepFile | editPlan | advancedElicitationTask | partyModeWorkflow |
|---|---|---|---|---|---|
| e-02-discover-edits | Discover what user wants to change about the agent | ./e-04-type-metadata.md | {bmb_creations_output_folder}/edit-plan-{agent-name}.md | {project-root}/_bmad/core/workflows/advanced-elicitation/workflow.xml | {project-root}/_bmad/core/workflows/party-mode/workflow.md |
Edit Step 2: Discover Edits
STEP GOAL:
Conduct targeted discovery to understand exactly what the user wants to change about their agent. Document all requested edits in structured format.
MANDATORY EXECUTION RULES (READ FIRST):
Universal Rules:
- 🛑 NEVER assume what edits are needed - ask explicitly
- 📖 CRITICAL: Read the complete step file before taking any action
- 🔄 CRITICAL: Read editPlan first to understand agent context
- 📋 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 editor consultant who helps users clarify their modification goals
- ✅ 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 vision for improvements, together we define precise edits
- ✅ Maintain collaborative inquisitive tone throughout
Step-Specific Rules:
- 🎯 Focus only on discovering what to edit, not how to implement yet
- 🚫 FORBIDDEN to make any modifications in this step
- 💬 Approach: Ask probing questions to understand edit scope
- 📋 Ensure all edits are documented to edit plan before proceeding
EXECUTION PROTOCOLS:
- 🎯 Guide conversation to uncover all desired changes
- 📊 Categorize edits by component (persona, commands, metadata, etc.)
- 💾 Document all edits to edit plan
- 🚫 FORBIDDEN to proceed without confirming all edits are captured
CONTEXT BOUNDARIES:
- Available context: editPlan with agent snapshot from previous step
- Focus: Discover what changes user wants to make
- Limits: Discovery and documentation only, no implementation
- Dependencies: Agent must be loaded in editPlan
MANDATORY SEQUENCE
CRITICAL: Follow this sequence exactly. Do not skip, reorder, or improvise unless user explicitly requests a change.
1. Read Edit Plan Context
Load the editPlan file first:
Read {editPlan} to understand the current agent structure and context.
2. Present Edit Categories
Guide the user through potential edit areas:
"What would you like to change about {agent-name}?
I can help you modify:
[P]ersona - Role, identity, communication style, principles [C]ommands - Add, remove, or modify commands and menu structure [M]etadata - Name, description, version, tags, category [A]ctions - Critical actions and activation behaviors [T]ype - Convert between Simple/Expert/Module types [O]ther - Configuration, capabilities, system context
Which areas would you like to edit? (You can select multiple)"
3. Deep Dive Discovery
For each selected category, ask targeted questions:
If Persona selected:
- "What aspect of the persona needs change?"
- "Should the role be more specific or expanded?"
- "Is the communication style hitting the right tone?"
- "Do the principles need refinement?"
If Commands selected:
- "Do you want to add new commands, remove existing ones, or modify?"
- "Are current command names and descriptions clear?"
- "Should command steps be adjusted?"
- "Is the menu structure working well?"
If Metadata selected:
- "What metadata fields need updating?"
- "Is the description accurate and compelling?"
- "Should version be bumped?"
- "Are tags still relevant?"
If Actions selected:
- "What critical actions need modification?"
- "Should new activation behaviors be added?"
- "Are current actions executing as expected?"
If Type conversion selected:
- "What type are you converting from/to?"
- "What's driving this conversion?"
- "Are you aware of the implications (e.g., Expert needs sidecar)?"
4. Document Edits to Plan
After discovery, append to editPlan:
## Edits Planned
### Persona Edits
- [ ] {edit description}
- [ ] {edit description}
### Command Edits
- [ ] {edit description}
- [ ] {edit description}
### Metadata Edits
- [ ] {edit description}
- [ ] {edit description}
### Critical Action Edits
- [ ] {edit description}
- [ ] {edit description}
### Type Conversion
- [ ] {from: X, to: Y, rationale: ...}
### Other Edits
- [ ] {edit description}
Present summary for confirmation:
"Here's what I heard you want to change:
{Summarize all edits in clear bulleted list}
Did I capture everything? Any edits to add, remove, or clarify?"
5. Present MENU OPTIONS
Display: "Select an Option: [A] Advanced Elicitation [P] Party Mode [C] Continue to Validation"
Menu Handling Logic:
- IF A: Execute {advancedElicitationTask}, and when finished redisplay the menu
- IF P: Execute {partyModeWorkflow}, and when finished redisplay the menu
- IF C: Save edits 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 [all edits documented and confirmed by user], will you then load and read fully {nextStepFile} to execute and checks.
🚨 SYSTEM SUCCESS/FAILURE METRICS
✅ SUCCESS:
- All desired edits discovered and documented
- Edits categorized by component type
- User confirmed edit list is complete
- Edit plan updated with structured edits
❌ SYSTEM FAILURE:
- Proceeding without documenting edits
- Missing edits that user mentioned
- Unclear or ambiguous edit descriptions
- User not given opportunity to review/edit list
Master Rule: Skipping steps, optimizing sequences, or not following exact instructions is FORBIDDEN and constitutes SYSTEM FAILURE.