brainstorming

name	brainstorming
description	You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.

Brainstorming Ideas Into Designs

Help turn ideas into fully formed designs and specs through natural collaborative dialogue.

Start by understanding the current project context, then ask questions one at a time to refine the idea. Once you understand what you're building, present the design and get user approval.

Do NOT invoke any implementation skill, write any code, scaffold any project, or take any implementation action until you have presented a design and the user has approved it. This holds for every feature, refactor, and bug fix, however simple it looks — with one carve-out, below, that is about size rather than about how simple something feels.

Anti-Pattern: "This Is Too Simple To Need A Design"

A todo list, a single-function utility, a settings screen: all of them go through this process. "Simple" projects are where unexamined assumptions cause the most wasted work. The design can be short — a few sentences for a genuinely small feature — but you MUST present it and get approval.

The failure this guards against is rationalizing: deciding mid-task that the thing you already understand does not need a design. If you are arguing for the exception, you are the case it was written for.

The one carve-out: mechanically trivial changes

A change is trivial when its shape is decided before you start and nothing about it is a judgment call: a typo, a mechanical rename, a formatting fix, a one-line config value, a broken link. No design exists to present — there is no alternative approach, no trade-off, nothing a user could approve differently.

For those, skip this skill and the chain: write the reason into .specify/trivial (one line) and make the edit. The phase-gate allows up to 10 lines while that marker is fresh and echoes the reason back. Say in your response that you took the trivial path and why.

The moment the change acquires a decision — an interface, a name others depend on, behavior, more than a handful of lines — it stopped being trivial and the gate above applies again. doubt-driven-development draws the same line for the same reason.

Checklist

You MUST create a task for each of these items and complete them in order:

1. Set the depth dial — default (conversational, the default) or deep (opt-in framed divergence). See ## Depth dial.
2. Explore project context — start from what SessionStart already injected: the product brief, the constitution, docs/STATE.md, and the OKF concept map (name, type, path, description of every documented concept). Those cost zero tool calls and already sit in your window; rediscovering them by reading files is waste. Then read only what is missing for THIS idea — the concepts the map points at, the code the feature touches, recent commits in that area.
3. Ask clarifying questions — one at a time, understand purpose/constraints/success criteria
4. Explore approaches — default: propose 2-3 inline with trade-offs and your recommendation. deep: dispatch framed divergence, then converge into 2-3 (see ## Depth dial).
5. Present design — in sections scaled to their complexity, get user approval after each section
6. Write design doc — copy .specify/templates/brainstorm-template.md to specs/<feature>/brainstorm.md and fill each section
7. Adversarial gate — run doubt-driven-development against the written design before handoff; fix actionable findings (see below)
8. User reviews written spec — ask user to review the spec file before proceeding
9. Hand off — set .specify/state to the feature name and hand off to prd-writer

Process Flow

digraph brainstorming {
    "Set depth dial\n(default | deep)" [shape=box];
    "Explore project context" [shape=box];
    "Ask clarifying questions" [shape=box];
    "Explore approaches\n(default: inline | deep: dispatch)" [shape=box];
    "Present design sections" [shape=box];
    "User approves design?" [shape=diamond];
    "Write design doc\n(from template)" [shape=box];
    "Adversarial gate\n(doubt-driven-development)" [shape=box];
    "User reviews spec?" [shape=diamond];
    "Hand off to prd-writer" [shape=doublecircle];

    "Set depth dial\n(default | deep)" -> "Explore project context";
    "Explore project context" -> "Ask clarifying questions";
    "Ask clarifying questions" -> "Explore approaches\n(default: inline | deep: dispatch)";
    "Explore approaches\n(default: inline | deep: dispatch)" -> "Present design sections";
    "Present design sections" -> "User approves design?";
    "User approves design?" -> "Present design sections" [label="no, revise"];
    "User approves design?" -> "Write design doc\n(from template)" [label="yes"];
    "Write design doc\n(from template)" -> "Adversarial gate\n(doubt-driven-development)";
    "Adversarial gate\n(doubt-driven-development)" -> "Write design doc\n(from template)" [label="actionable findings"];
    "Adversarial gate\n(doubt-driven-development)" -> "User reviews spec?" [label="clean"];
    "User reviews spec?" -> "Write design doc\n(from template)" [label="changes requested"];
    "User reviews spec?" -> "Hand off to prd-writer" [label="approved"];
}

The terminal state is handing off to prd-writer. Do NOT invoke any other implementation skill. The ONLY skill you invoke after brainstorming is prd-writer.

The Process

Understanding the idea:

• Check out the current project state first (files, docs, recent commits)
• Before asking detailed questions, assess scope: if the request describes multiple independent subsystems (e.g., "build a platform with chat, file storage, billing, and analytics"), flag this immediately. Don't spend questions refining details of a project that needs to be decomposed first.
• If the project is too large for a single spec, help the user decompose into sub-projects: what are the independent pieces, how do they relate, what order should they be built? Then brainstorm the first sub-project through the normal design flow. Each sub-project gets its own spec → plan → implementation cycle.
• For appropriately-scoped projects, ask questions one at a time to refine the idea
• Prefer multiple choice questions when possible, but open-ended is fine too
• Only one question per message - if a topic needs more exploration, break it into multiple questions
• Focus on understanding: purpose, constraints, success criteria

Exploring approaches:

• default: propose 2-3 different approaches with trade-offs, present them conversationally, lead with your recommended option and explain why.
• deep (opt-in): when the design has real forks and the user opted in, run the framed divergence in ## Depth dial instead — the 2-3 approaches you present are the critic's survivors, not options you anchored inline.
• Either way, the approaches (and which won, and why) land in the Approaches Considered section of the brainstorm doc.

Presenting the design:

• Once you believe you understand what you're building, present the design
• Scale each section to its complexity: a few sentences if straightforward, up to 200-300 words if nuanced
• Ask after each section whether it looks right so far
• Cover: architecture, components, data flow, error handling, testing
• Be ready to go back and clarify if something doesn't make sense

Design for isolation and clarity:

• Break the system into smaller units that each have one clear purpose, communicate through well-defined interfaces, and can be understood and tested independently
• For each unit, you should be able to answer: what does it do, how do you use it, and what does it depend on?
• Can someone understand what a unit does without reading its internals? Can you change the internals without breaking consumers? If not, the boundaries need work.
• Smaller, well-bounded units are also easier for you to work with - you reason better about code you can hold in context at once, and your edits are more reliable when files are focused. When a file grows large, that's often a signal that it's doing too much.

Working in existing codebases:

• Explore the current structure before proposing changes. Follow existing patterns.
• Where existing code has problems that affect the work (e.g., a file that's grown too large, unclear boundaries, tangled responsibilities), include targeted improvements as part of the design - the way a good developer improves code they're working in.
• Don't propose unrelated refactoring. Stay focused on what serves the current goal.

After the Design

Documentation:

• Copy .specify/templates/brainstorm-template.md to specs/<feature>/brainstorm.md, where <feature> is the next NNN-kebab-name ordinal (same numbering scheme prd-writer uses: highest existing specs/* ordinal + 1, zero-padded to 3 digits). Fill each section (Understanding / Investigation / Approaches Considered / Open Decisions / Solution Outline) and the status/feature/date frontmatter. This template is the durable capture of the reasoning trail — the exploration, the trade-offs, the open points — upstream of spec.md/plan.md. No code, only intent and contracts.
• Set .specify/state (single line, no trailing content) to NNN-kebab-name so it becomes the active feature for prd-writer.
• Do NOT commit the design document — this repo commits at feature/milestone boundaries with explicit human approval, not per artifact.

Adversarial Gate: After writing the design doc, run doubt-driven-development against it before handing off — a fresh-context reviewer that tries to disprove the design while course-correction is still cheap. This replaces the old cosmetic self-review: the point is an attack, not a proofread.

• EXTRACT the design's contract (the Outline + the closed decisions) and hand the reviewer that artifact without your reasoning or your conclusion.
• DOUBT: the adversarial brief is "here is a proposed design and its constraints — find where it is wrong, unsafe, under-scoped, or built on an unstated assumption."
• RECONCILE each finding: actionable → fix the doc; contract misread → tighten the EXTRACT; trade-off → record it in Open Decisions (or as closed, with the rationale); noise → discard. Bounded at 3 cycles; escalate rather than grind a fourth.
• Fold the old quick checks into this pass: placeholder scan (no "TBD"/"TODO"/vague requirements), internal consistency, scope (single plan or decompose?), and ambiguity (each requirement interpretable one way). A finding here is just another actionable.

Skip the full adversarial dispatch only for a truly trivial, reversible design (default on a one-liner) — there, the fold-in checks alone suffice; say so explicitly.

User Review Gate: After the adversarial gate comes back clean, ask the user to review the written spec before proceeding:

"Design written to specs/<feature>/brainstorm.md. Please review it and let me know if you want to make any changes before we hand off to prd-writer."

Wait for the user's response. If they request changes, make them and re-run the spec review loop. Only proceed once the user approves.

Hand-off:

• Hand off to the prd-writer skill to turn the approved design into the feature's PRD.
• Do NOT invoke any other skill. prd-writer is the next step.

Depth dial

Set once, up front. The dial scales the process, not just the doc length — small features stay conversational, real design forks unlock divergence.

default. The conversational flow: explore context, ask one question at a time, propose 2-3 approaches inline, present the design. This is the right choice for most features, and for anything the off-signals already wave through (a typo, a rename, a trivial config, a 1-file change) — those barely need a dial at all. Do not reach for deep by default; it costs subagents.

deep (opt-in). For a design with genuine forks — an architecture choice under real uncertainty, an API surface, a naming/modeling decision where the first idea anchors all the others. Offer it; do not impose it. When the user opts in, run framed divergence → convergence:

1. Diverge. Dispatch a small fixed set of isolated idea-agents (via dispatching-parallel-agents, cheapest capable model — divergence is generative, not precise). Each agent sees the problem plus one cognitive frame and is told to generate, not evaluate. Agents do not see each other — no cross-anchoring. A lean, generic frame set (not codebase-specific): skeptical-user, maintenance-6-months-later, worst-case-load, first-principles, simplest-path. Pick 3-4 relevant to the fork.
2. Converge (critic). A single fresh pass over all branches: score each idea (novelty / viability / fit), flag traps (seductive-but-broken, with the reason), cluster by underlying angle, and deepen the top 2-3 survivors into approaches with risks and first steps.
3. The survivors become the 2-3 approaches you present, and land in Approaches Considered with the trade-off trail.

deep is the framed-divergence primitive borrowed from the adhd pattern, built entirely on keel's own dispatch — not a separate framework. It is a rare, opt-in mode, never the default.

Key Principles

• Ask one question at a time - Don't overwhelm with multiple questions
• Multiple choice preferred - Easier to answer than open-ended when possible
• YAGNI ruthlessly - Remove unnecessary features from all designs
• Explore alternatives - Always propose 2-3 approaches before settling
• Incremental validation - Present design, get approval before moving on
• Be flexible - Go back and clarify when something doesn't make sense

Visual Companion (optional)

For genuinely visual questions — a real mockup, layout, or diagram question, not merely a UI topic — a browser-based visual companion tool can be offered just-in-time (its own message, not upfront) so options can be shown rather than described. If no visual question ever arises, never offer it, and text-only dialogue is the default.