You MUST use this before any creative work — features, components, behavior changes. Explores intent, requirements, and design before implementation.
when_to_use
[feedforward] Triggered before any creative or implementation work to explore intent and design.
disable-model-invocation
true
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 clarify decisions in frontier rounds: ask every currently-unblocked question together, each with a recommended answer, then recompute the frontier after the user responds. 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 applies to EVERY project regardless of perceived simplicity.
Anti-Pattern: "This Is Too Simple To Need A Design"
Every project goes through this process. A todo list, a single-function utility, a config change — all of them. "Simple" projects are where unexamined assumptions cause the most wasted work. The design can be short (a few sentences for truly simple projects), but you MUST present it and get approval.
Checklist
You MUST create a task for each of these items and complete them in order:
Knowledge-base retrieval convention: read docs/agent-harness/index.md first, then follow the relevant topic index; do not use global **/*.md globs
Map decision tree — optional, complex tasks only (3+ decision dimensions): sketch the key decision dependencies before questioning
Ask clarifying questions — work the current frontier in rounds; ask all unblocked decision questions together, each with a recommended answer and reason
When domain terms crystallize (user defines a concept, or you propose a precise term to replace fuzzy language), invoke agent-harness:domain-modeling to update CONTEXT.md inline. If CONTEXT.md doesn't exist yet, the skill creates it lazily. Spec output should use CONTEXT.md vocabulary and include a domain_terms field in frontmatter listing the core terms.
Propose 2-3 approaches — with trade-offs and your recommendation
Present design — in sections scaled to their complexity, get user approval after each section
Write design doc — save to docs/agent-harness/specs/YYYY-MM-DD-<topic>-design.md (check if target directory is gitignored before committing; if so, inform user and save anyway)
Spec self-review — quick inline check for placeholders, contradictions, ambiguity, scope (see below)
User reviews written spec — ask user to review the spec file before proceeding
Transition to implementation — invoke writing-plans skill to create implementation plan
The terminal state is invoking writing-plans. Between design approval and writing-plans, you MAY optionally invoke agent-harness:gate-driven-test-design when the user asks for test case generation or the feature carries non-trivial behavior/contract/regression risk. The ONLY skills you invoke after brainstorming are gate-driven-test-design (optional) and writing-plans (required). Do NOT invoke frontend-design, mcp-builder, or any other implementation skill.
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.
For large tasks (full-stack project, multiple apps, backend + frontend + AI, or likely >8 implementation tasks), do large-task decomposition first: present a directory-level execution map with the proposed sub-plans and confirmation gates. Do not write a monolithic spec/plan. Default split: infrastructure / backend / frontend / design-polish, adjusted to the project.
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 complex tasks with 3+ decision dimensions, map the decision tree before questioning: identify what must be decided first and which later questions depend on each branch. Simple tasks can keep this tree implicit.
Ask clarifying questions:
Build the current frontier: every decision whose prerequisites are already settled and can be answered now
Ask the whole frontier in one round: number each question and include a recommended answer with the reason
Wait for the user's answers, then recompute the frontier; settled decisions unblock dependent questions for the next round
If question B depends on question A's answer, B belongs to a later round, not the current one
Do not force 3+ questions per round: when the real frontier has only one question, ask one question
Recommended answers are defaults, not constraints: the user may accept, modify, or reject them
Fact-checking rules:
Separate facts from decisions
Facts are things you can look up: project file structure, existing API endpoints, config files, codebase patterns
Decisions require human judgment: technology choices, business logic, product trade-offs, design preferences
Facts are your job: use tools or dispatch subagents to check them; never ask the user for facts you can inspect
Do not block unnecessarily: while a fact-finding subagent runs, keep asking frontier questions that do not depend on that fact
Only questions downstream of an unresolved fact wait for the fact-finding result
Exploring approaches:
Propose 2-3 different approaches with trade-offs
Present options conversationally with your recommendation and reasoning
Lead with your recommended option and explain why
YAGNI ruthlessly — remove unnecessary features from every approach and design
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
Keep units isolated and understandable: each component should have one clear purpose, explicit dependencies, and boundaries a reader can understand without reading internals
Be ready to go back and clarify if something doesn't make sense
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:
Write the validated design (spec) to docs/agent-harness/specs/YYYY-MM-DD-<topic>-design.md
(User preferences for spec location override this default)
Every spec document must start with YAML frontmatter before the title:
User Review Gate:
After the spec review loop passes, ask the user to review the written spec before proceeding:
"Spec written and committed to <path>. Please review it and let me know if you want to make any changes before we start writing out the implementation plan."
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.
Sprint Contract:
After spec approval, before invoking writing-plans, use agent-harness:sprint-contract to negotiate explicit Definition of Done. This prevents the common failure mode of "completed but not what was expected."
Skip sprint contract only for changes that meet ALL of these criteria:
Scope: single file, < 15 lines changed
Risk: no architecture decisions, no new API surface, no behavioral change to existing logic
Clarity: the user's request is unambiguous and the implementation path is obvious
Examples of valid skips: typo fixes, pure documentation changes, renaming a variable, updating a constant value.
If unsure whether a change qualifies, default to running sprint contract.
Implementation:
Invoke the writing-plans skill to create a detailed implementation plan
Do NOT invoke any other skill. writing-plans is the next step.
Clarification Loop Circuit-Breaker (issue #83)
If the user rejects your proposed options 3 times in a row (clear rejection
signals — "不对" / "不行" / "重新" / "no" / "not what I meant" / "that's not
it" / "try again" / a hesitant "嗯..." followed by a different question / any
response that says "this isn't what I want, try something else" regardless of
language), stop listing more options. Listing more variants of the same
shape does not converge — it inflates context with zero-output turns (hack
session 90b1b2fd hit 45.6% no-tool turns this way).
Judge by intent, not by keyword matching — a user can reject without using any
of the example phrases. If you're unsure whether something was a rejection,
ask directly ("is this a no?") rather than treating an ambiguous turn as
neither yes nor no.
Switch strategy immediately:
Stop generating option lists.
Ask one open-ended outcome question: "能描述一下你最终想看到的结果是什么
样子吗？不用管可行性。" / "Describe the end result you want to see, ignoring
feasibility for now."
If the user still can't describe it, recommend handoff:
agent-harness:office-hours to re-align on goals, or
pause and ask the user to gather more context before continuing.
Only resume option-listing once the user has described the desired outcome
in their own words.
Six Forcing Questions (Product Ideas)
When brainstorming a new product idea or major new feature (not bug fixes or small improvements), use these six forcing questions to validate the idea before diving into design. These questions expose assumptions and prevent building things nobody wants.
When to use: User says "I have an idea", "is this worth building", "help me think through this", or describes a new product/feature concept.
How to use: Ask these during the frontier questioning phase when they are decision questions. Do not use them to ask for facts the agent can inspect. Not every question needs a long answer — some can be quick. But each must be addressed.
#
Question
What It Exposes
1
Demand Reality — Who specifically has this problem, and how do you know? Have you talked to them?
Prevents building for imagined users
2
Status Quo — How do people solve this problem today? What's wrong with that?
Reveals if the pain is real and if alternatives exist
3
Desperate Specificity — Who would be desperate for this? Describe them precisely.
Forces narrow focus vs. "everyone would want this"
4
Narrowest Wedge — What's the smallest possible version that solves the core problem?
Prevents scope creep before validation
5
Observation — What have you personally observed that makes you believe this is needed?
Distinguishes insight from assumption
6
Future-Fit — If this works, what does it grow into? If it fails, what did you learn?
Tests strategic thinking and learning mindset
After the six questions: If the idea survives scrutiny, proceed to normal brainstorming (approaches, design, spec). If the questions reveal weak foundations, help the user either:
Pivot to a stronger version of the idea
Identify what research/validation is needed first
Decide to shelve the idea
Skip these questions for:
Bug fixes
Small improvements to existing features
Technical refactoring
Implementation of already-validated requirements
Capture Learnings
After design approval, if you discovered something worth remembering:
User stated a strong preference (naming, style, architecture approach)
Discovered a project convention not documented elsewhere
Made an architectural decision with non-obvious rationale
Record it using session-learnings skill so future sessions respect these decisions.
