Files
brachnha-insight/_bmad/bmb/workflows/agent/steps-c/step-02-discovery.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

5.4 KiB

name, description, nextStepFile, agentPlan, brainstormContext, advancedElicitationTask, partyModeWorkflow
name description nextStepFile agentPlan brainstormContext advancedElicitationTask partyModeWorkflow
step-02-discovery Discover what user wants holistically ./step-03-type-metadata.md {bmb_creations_output_folder}/agent-plan-{agent_name}.md ../data/brainstorm-context.md {project-root}/_bmad/core/workflows/advanced-elicitation/workflow.xml {project-root}/_bmad/core/workflows/party-mode/workflow.md

STEP GOAL

Conduct holistic discovery of what the user wants to create, documenting a comprehensive agent plan that serves as the single source of truth for all subsequent workflow steps. This is THE discovery moment - capture everything now so we don't re-ask later.

MANDATORY EXECUTION RULES

  1. ONE-TIME DISCOVERY: This is the only discovery step. Capture everything now.
  2. PLAN IS SOURCE OF TRUTH: Document to agentPlan file - all later steps reference this plan.
  3. NO RE-ASKING: Later steps MUST read from plan, not re-ask questions.
  4. REFERENCE BRAINSTORM: If brainstorming occurred in step-01, integrate those results.
  5. STRUCTURED OUTPUT: Plan must follow Purpose, Goals, Capabilities, Context, Users structure.
  6. LANGUAGE ALIGNMENT: Continue using {language} if configured in step-01.

EXECUTION PROTOCOLS

Protocol 1: Check for Previous Context

Before starting discovery:

  • Check if brainstormContext file exists
  • If yes, read and reference those results
  • Integrate brainstorming insights into conversation naturally

Protocol 2: Discovery Conversation

Guide the user through holistic discovery covering:

  1. Purpose: What problem does this agent solve? Why does it need to exist?
  2. Goals: What should this agent accomplish? What defines success?
  3. Capabilities: What specific abilities should it have? What tools/skills?
  4. Context: Where will it be used? What's the environment/setting?
  5. Users: Who will use this agent? What's their skill level?

Use conversational exploration:

  • Ask open-ended questions
  • Probe deeper on important aspects
  • Validate understanding
  • Uncover implicit requirements

Protocol 3: Documentation

Document findings to agentPlan file using this structure:

# Agent Plan: {agent_name}

## Purpose
[Clear, concise statement of why this agent exists]

## Goals
- [Primary goal 1]
- [Primary goal 2]
- [Secondary goals as needed]

## Capabilities
- [Core capability 1]
- [Core capability 2]
- [Additional capabilities with tools/skills]

## Context
[Deployment environment, use cases, constraints]

## Users
- [Target audience description]
- [Skill level assumptions]
- [Usage patterns]

Protocol 4: Completion Menu

After documentation, present menu:

[A]dvanced Discovery - Invoke advanced-elicitation task for deeper exploration [P]arty Mode - Invoke party-mode workflow for creative ideation [C]ontinue - Proceed to next step (type-metadata)

CONTEXT BOUNDARIES

DISCOVER:

  • Agent purpose and problem domain
  • Success metrics and goals
  • Required capabilities and tools
  • Usage context and environment
  • Target users and skill levels

DO NOT DISCOVER:

  • Technical implementation details (later steps)
  • Exact persona traits (next step)
  • Command structures (later step)
  • Name/branding (later step)
  • Validation criteria (later step)

KEEP IN SCOPE:

  • Holistic understanding of what to build
  • Clear articulation of value proposition
  • Comprehensive capability mapping

MANDATORY SEQUENCE

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

  1. Load Previous Context

    • Check for brainstormContext file
    • Read if exists, note integration points
  2. Start Discovery Conversation

    • Reference brainstorming results if available
    • "Let's discover what you want to create..."
    • Explore purpose, goals, capabilities, context, users
  3. Document Plan

    • Create agentPlan file
    • Structure with Purpose, Goals, Capabilities, Context, Users
    • Ensure completeness and clarity
  4. Present Completion Menu

    • Show [A]dvanced Discovery option
    • Show [P]arty Mode option
    • Show [C]ontinue to next step
    • Await user selection
  5. Handle Menu Choice

    • If A: Invoke advanced-elicitation task, then re-document
    • If P: Invoke party-mode workflow, then re-document
    • If C: Proceed to step-03-type-metadata

CRITICAL STEP COMPLETION NOTE

THIS STEP IS COMPLETE WHEN:

  • agentPlan file exists with complete structure
  • All five sections (Purpose, Goals, Capabilities, Context, Users) populated
  • User confirms accuracy via menu selection
  • Either continuing to next step or invoking optional workflows

BEFORE PROCEEDING:

  • Verify plan file is readable
  • Ensure content is sufficient for subsequent steps
  • Confirm user is satisfied with discoveries

SUCCESS METRICS

SUCCESS:

  • agentPlan file created with all required sections
  • User has provided clear, actionable requirements
  • Plan contains sufficient detail for persona, commands, and name steps
  • User explicitly chooses to continue or invokes optional workflow

FAILURE:

  • Unable to extract coherent purpose or goals
  • User cannot articulate basic requirements
  • Plan sections remain incomplete or vague
  • User requests restart

RECOVERY:

  • If requirements unclear, use advanced-elicitation task
  • If user stuck, offer party-mode for creative exploration
  • If still unclear, suggest revisiting brainstorming step