Skills
Shape a marketing outcome with ak:brainstorm
Use ak:brainstorm to turn incomplete intent into an accepted outcome, constraints, non-goals, and observable acceptance criteria.
Use ak:brainstorm before multi-step marketing delivery when the desired
outcome, boundaries, or approach is not yet settled. The Skill converts an
incomplete request into a bounded contract, inspects relevant evidence, and
compares up to three viable directions when a real choice exists. It does not
execute the selected direction.
Choose ak:brainstorm before delivery
Use ak:brainstorm when
- A campaign, content program, funnel change, or launch has a goal but not a concrete deliverable and acceptance evidence.
- Several approaches could satisfy the objective and their trade-offs matter.
- You need to separate the intended outcome from assumptions about current audience, channel, product, or project behavior.
- A diagnosed problem has multiple cause-aligned solutions.
- A decision needs a durable HTML brief for review before planning.
Choose another workflow when
- You need a direct technical answer without a design loop. Use
ak:ask. - The direction is accepted and you need a durable implementation or CRO plan.
Use
ak:plan. - A concrete bug has not been diagnosed. Use
ak:debugorak:fix; do not brainstorm fixes from symptoms. - The task is already clear and only needs domain execution. Use the relevant Marketing Skill, keeping publishing, spend, outreach, and account changes as separate approvals.
Prepare the opening contract
Bring the best available evidence and be ready to confirm four fields:
| Field | What to provide |
|---|---|
| Outcome | The user-visible, audience-visible, or operational end state. |
| Constraints | Brand, privacy, compliance, channel, timing, compatibility, budget, and ownership boundaries. |
| Non-goals | Nearby work the delivery must not absorb. |
| Acceptance criteria | Observable evidence that would prove the outcome is ready for the next stage. |
If an accepted brief or plan already contains these fields, point the Skill to it. The workflow should reuse settled decisions and ask only about a material gap.
| Runtime | Invocation | Availability boundary |
|---|---|---|
| Claude Code | /ak:brainstorm ... | Native and explicit plugin delivery are supported. |
| Cursor | /ak:brainstorm ... | Projected as a slash-invoked Skill; broader parity is not implied. |
| Codex | $ak:brainstorm ... | Uses native Codex Skill discovery; surrounding automation differs by runtime. |
Run the Skill
/ak:brainstorm "Define a launch outcome for the new team workspace. Use the supplied interview notes, exclude paid media and outbound outreach, and compare at most three directions"/ak:brainstorm "Define a launch outcome for the new team workspace. Use the supplied interview notes, exclude paid media and outbound outreach, and compare at most three directions"$ak:brainstorm "Define a launch outcome for the new team workspace. Use the supplied interview notes, exclude paid media and outbound outreach, and compare at most three directions"Choose optional behavior deliberately
| Option | Effect | Boundary |
|---|---|---|
| No flag | Returns the accepted contract, compared approaches when needed, recommendation, evidence, and risks while preserving the full requested scope | Adds nothing unrequested and does not implement the direction |
--ask | Runs a bounded evidence-first interview before finalizing the contract | User-invoked only; never forwarded downstream. Headless sessions apply recommended defaults for reversible choices and hard-stop on security-critical gaps. |
--html | Writes a self-contained brainstorm.html in the repository's configured report location | This is a workspace mutation; it does not publish the file |
--report | Writes a timestamped Markdown report with the accepted contract, options, recommendation, and unresolved questions | Uses the active plan's reports directory when available, otherwise plans/reports/ |
--advice | Adds kongming advisory checkpoints after decisions, when stuck, and before high-stakes choices | Advice cannot approve gates, write deliverables, or expand authority |
--ultra | Runs a best-of-5 verifier pass over one shared evidence packet and rubric | Five read-only candidate brainstorms run independently; a verifier selects one winning contract unchanged or rejects all candidates |
--yagni | Challenges and cuts scope not needed for the stated outcome | KISS and DRY still apply; explicit safety and acceptance requirements remain in scope |
When --ultra is selected, the Skill keeps candidate work read-only, requires
five usable candidates with one bounded re-dispatch if needed, and records a
short ranking appendix. It requires parallel multi-candidate dispatch and stops
when the runtime cannot provide it. It is inspired by LLM-as-a-Verifier, but
does not claim logprob or tournament mechanics. With --report, the Markdown
report records the winning candidate and ranking appendix.
When --html is selected, the brief includes the four contract fields,
approach trade-offs, recommendation, unresolved risks, and an implementation
workflow diagram. For UI/UX topics it also calls for annotated mockups. The
file keeps CSS and JavaScript inline, but default AntV panels may contact
jsDelivr. For offline or disclosure-sensitive work, use --no-antv or
--no-editorial-visuals. When visual preferences enable them, approach
comparisons prefer a diagram-design Quadrant and effort or impact tiles prefer
AntV Infographic CandyCardLite or CompactCard. Disable these additions with
--no-antv,
--no-diagram-design, or --no-editorial-visuals.
With --advice, kongming uses the strongest advisory tier for the detected
host: Fable 5 for a Claude Code subscription, gpt-6-astra with low reasoning
for Codex, or claude-fable-5-high for Cursor. Other hosts use the strongest
available session model and report that they could not pin Fable or Astra.
Understand the workflow stages
- Capture or reuse the contract. The Skill identifies the outcome, constraints, non-goals, and acceptance criteria.
- Inspect relevant evidence. It reads the smallest useful project, content, test, or live-state surface before claiming an approach is feasible.
- Route bugs through diagnosis. It captures the expected repaired behavior, then scouts and proves the root cause before discussing solutions.
- Compare meaningful options. When a real design choice exists, it presents up to three viable approaches and their trade-offs.
- Recommend a sufficient direction. It applies KISS and DRY, preserves the
full requested scope, adds nothing unrequested, and records remaining
uncertainty. With
--yagni, it also challenges unnecessary scope. - Hand off without executing. It passes the contract, chosen direction, evidence, and unresolved risks to a planning or delivery workflow.
A clear request does not require a ceremonial interview. A missing answer should be requested only when it would materially change the result, safety boundary, or public contract and cannot be discovered from evidence.
Keep decisions and mutations separate
A selected direction is not execution approval
ak:brainstorm does not publish content, change a campaign or provider
account, contact an audience, spend budget, deploy, or implement the chosen
direction. --advice does not change that boundary. --html authorizes only
the local brief described above.
If the workflow inspects analytics, customer research, or live platforms, limit it to approved data and current access. Do not send credentials, secrets, personal data, or confidential audience records to an external service without separate authorization and an acceptable data-handling path.
Review the output and evidence
A useful handoff should include:
- The accepted outcome, constraints, non-goals, and acceptance criteria.
- The current evidence inspected and any gaps or uncertainty.
- Up to three viable approaches when the choice is material.
- A recommended direction with explicit trade-offs.
- Risks, unresolved decisions, and the next owning workflow.
- A local
brainstorm.htmlonly when--htmlwas requested. - A timestamped Markdown report only when
--reportwas requested.
The result is ready for planning when the contract is concrete and the chosen direction is supported by the available evidence. It is not proof that the campaign, content, or product change will produce a business result.
Troubleshoot the brainstorm
| Symptom | Safe next step |
|---|---|
The runtime cannot find ak:brainstorm | Confirm target and scope, restart the runtime, then follow Runtime cannot find a Skill or Agent. |
| The Skill asks many low-value questions | Provide the four contract fields and point to already accepted evidence or decisions. |
| Options are generic | Supply audience, offer, channel, brand, privacy, timing, and non-goal constraints plus the evidence the Skill may inspect. |
| The workflow proposes a fix before diagnosis | Stop and route the concrete failure through ak:debug or ak:fix. |
brainstorm.html is missing | Confirm --html was passed and that a configured report location is writable; report the limitation instead of writing elsewhere. |
| The Markdown report is missing | Confirm --report was passed and that the active plan or standalone reports directory is writable. |
| The Skill starts implementation | Stop at the handoff and require a separate approved plan or domain workflow. |
Know the current limits
- Brainstorming improves decision structure; it cannot guarantee demand, conversion, revenue, campaign performance, or platform support.
- Evidence quality is limited by the inputs, project access, tools, and current provider state available in the session.
--htmlcan represent a proposed experience, but it is not a deployed page, tested campaign, or approved creative asset.
Continue with ak:plan for a durable plan or return to the
Marketing Kit overview to choose a domain workflow.
Restate the previous answer plainly with ak:bro
Make the immediately previous assistant message shorter and easier to understand without changing its meaning or taking action.
Make an outcome-focused priority call with ak:sowat
Review recent implementation and related issues like a product owner, correct weak priorities, and recommend up to three evidence-backed next steps.