brainstorming-workflows

Brainstorming Workflows

Overview

Brainstorming workflows IS understanding the human before designing the system.

Use role templates to quickly identify the user's context, then explore their specific workflows one question at a time. Don't assume the user is a developer — they may use the agent system for project management, content creation, data analysis, or other work.

Core principle: The agent system must serve the user's actual workflows, not an imagined ideal.

Violating the letter of the rules is violating the spirit of the rules.

Routing

Pattern: Chain Handoff: user-confirmation Next: planning-agent-systems Chain: main

Task Initialization (MANDATORY)

Before ANY action, create task list using TaskCreate:

TaskCreate for EACH task below:
- Subject: "[brainstorming-workflows] Task N: <action>"
- ActiveForm: "<doing action>"

Tasks:

1. Import analysis findings (if available)
2. Role selection
3. Workflow exploration
4. Identify simplest viable approach
5. Produce workflow summary

Announce: "Created 5 tasks. Starting execution..."

Execution rules:

1. TaskUpdate status="in_progress" BEFORE starting each task
2. TaskUpdate status="completed" ONLY after verification passes
3. If task fails → stay in_progress, diagnose, retry
4. NEVER skip to next task until current is completed
5. At end, TaskList to confirm all completed

Task 1: Import Analysis Findings (if available)

Goal: If an analysis report exists, bring its findings into the conversation.

If analysis report path was provided:

1. Read the analysis report
2. Summarize critical and warning findings
3. Ask user: "分析發現以下弱點，是否要在這次一併修補？"
4. Present findings as a checklist for user to select
5. Record selected items → these become requirements in the workflow summary

If no analysis report: Skip to Task 2.

This allows skipping questions already answered by analysis. For example:

• Analysis found "no linting hook" → skip asking about code quality tools
• Analysis found "CLAUDE.md > 200 lines" → skip asking about constitution preferences

Verification: User has confirmed which findings to address (or no analysis exists).

Task 2: Role Selection

Goal: Identify the user's primary role to guide exploration.

Present the role table:

Role	Typical Workflows
A) Software Developer	coding, testing, code review, CI/CD, deployment
B) Project Manager	task tracking, reporting, scheduling, communication
C) Content Creator	writing, translation, publishing, social media
D) Data Analyst	data processing, visualization, reporting, automation
E) Operations / DevOps	monitoring, deployment, incident response, IaC
F) Custom	describe your role

Ask: "你的角色最接近哪一個？選擇字母即可。"

Verification: User has selected a role.

Task 3: Workflow Exploration

Goal: Explore the user's specific workflows one question at a time.

CRITICAL: Read references/role-templates.md for role-specific deep-dive questions.

Rules:

• One question at a time — never ask multiple questions in one message
• Skip questions answered by analysis — don't re-ask what we already know
• Multiple choice when possible — easier for user to answer
• Adapt to answers — if user reveals something unexpected, explore it
• 5-8 questions maximum — don't exhaust the user
• Ask about failures — "What have you tried before? What didn't work?" reveals more than "What do you want?"
• Ask for a walkthrough — "Walk me through the last time you did X, step by step" beats generic "What tools do you use?"

After workflow exploration, classify into two layers:

Layer 1 — Anthropic workflow patterns (how workflows execute): Read references/anthropic-patterns.md for the six patterns and their signal phrases from user answers.

Layer 2 — Skill routing patterns (how skills connect): Read references/routing-patterns.md for Tree/Chain/Node/Skill Steps classification.

Verification: Have enough information to map workflows to BOTH Anthropic patterns and skill routing patterns.

Task 4: Identify Simplest Viable Approach

Goal: Before producing the summary, challenge every workflow for simplicity.

Core question: "What is the simplest approach that works for each workflow?"

Anthropic's guidance: "Success in the LLM space isn't about building the most sophisticated system. It's about building the right level of complexity for your needs."

For each workflow, ask:

1. Can this be solved with a single CLAUDE.md instruction instead of a skill?
2. Can this be a rule instead of a hook?
3. Can this be a simple prompt chain instead of a multi-agent orchestration?
4. Does this need a custom skill, or does an existing skill/plugin already handle it?

Read references/anthropic-patterns.md for the complexity ladder (Levels 1-6). Prefer the lowest level that works.

Present the assessment to the user: Show each workflow with its proposed complexity level and ask if they agree. Users often accept over-engineering without questioning — push back gently.

Verification: Every workflow has an assigned complexity level, and the user has confirmed the approach.

Task 5: Produce Workflow Summary

Goal: Write structured summary to docs/agent-system/{timestamp}-workflows.md.

CRITICAL: Read references/summary-template.md for the full summary format.

Handoff: "工作流摘要完成。要繼續規劃 agent system 元件嗎？"

• If yes → invoke planning-agent-systems skill, pass workflow summary path

Verification: Summary written with all workflows mapped to components.

Red Flags - STOP

These thoughts mean you're rationalizing. STOP and reconsider:

Thought	Reality
"Obviously a developer"	PMs, analysts, creators all use agent systems. Ask.
"I know the workflows"	You know common workflows. Theirs may differ.
"Multiple questions saves time"	Multiple questions overwhelm. One at a time.
"Skip analysis"	Analysis findings prevent redundant questions. Use them.
"Summary is overhead"	Summary is the contract for planning. Essential.
"This needs a skill"	Most workflows need less than you think. Check the complexity ladder.
"Skip past failures"	Past failures are the highest-value context. Always ask.

Flowchart: Workflow Brainstorming

digraph brainstorm_workflows {
    rankdir=TB;

    start [label="Brainstorm\nworkflows", shape=doublecircle];
    analysis [label="Task 1: Import\nanalysis findings", shape=box];
    has_analysis [label="Analysis\nexists?", shape=diamond];
    confirm_fixes [label="User selects\nfindings to fix", shape=box];
    role [label="Task 2: Role\nselection", shape=box];
    explore [label="Task 3: Workflow\nexploration", shape=box];
    simplify [label="Task 4: Simplest\nviable approach", shape=box, style=filled, fillcolor="#ffffcc"];
    summary [label="Task 5: Produce\nworkflow summary", shape=box];
    handoff [label="Invoke\nplanning-agent-systems", shape=box];
    done [label="Brainstorm complete", shape=doublecircle];

    start -> analysis;
    analysis -> has_analysis;
    has_analysis -> confirm_fixes [label="yes"];
    has_analysis -> role [label="no"];
    confirm_fixes -> role;
    role -> explore;
    explore -> simplify;
    simplify -> summary;
    summary -> handoff [label="continue"];
    summary -> done [label="stop here"];
    handoff -> done;
}

References

• references/role-templates.md — Role-specific deep-dive questions and component mappings
• references/anthropic-patterns.md — Six Anthropic workflow patterns and complexity ladder
• references/summary-template.md — Workflow summary document format