AGENT SKILL

brainstorm

Brainstorm the intervention — ~10 candidates for a named problem in any domain, from today to a quarter-long bet. Use on "brainstorm interventions", "show me options". NOT for design variants.

License 许可
MIT
最近更新
2026-11-13
Stars
★ 2
Forks
0

01 前言 Front Matter

name
brainstorm
description
Brainstorm the intervention — ~10 candidates for a named problem in any domain, from today to a quarter-long bet. Use on "brainstorm interventions", "show me options". NOT for design variants.

02 资源文件 1

03 SKILL.md 正文

Brainstorm — Solution-Space Map (Brainstorm the Intervention)

This skill serves what neither the user nor you know yet, and surfaces both sides' unknowns.

When you and the user both seem to understand the request, do it without this method: if its result carries weight (it ships, is sent or is hard to undo), first read back in 2–3 lines the request as you understand it, the assumptions you are making and what outcome counts as done, as the whole reply, and do it once the user confirms; a self-evident request (fix a typo) needs no read-back. When the read-back exposes a mismatch, route by who does not know: the user → the teach-me skill; you → the interview skill; neither of you → the blindspot skill. A misread weighty request costs the work built on it; a read-back costs one exchange.

Jumping on the first solution leaves the rest of the solution space unknown. The user picks; you draw the space.

Iron Rules

  1. The output is a map of solutions, not a solution. It ends at the user's assembled selection.
  2. Spread candidates across time horizons and approaches, not 10 near-duplicates.

Procedure

  1. Get the problem statement and constraints (timeline, people, cost) from the text the user gave with the request or after the skill name and the conversation. If they are empty or vague, ask back once for a measurable statement: which metric, how bad, since when, for whom. One round only. If the answer is still vague or the user skips, generate the candidates anyway and state the assumed metric and constraints above them.

    Your own unknowns: say what you do not know about the user's situation, what you are assuming, and what you cannot check, and ask the user to fill or correct them. When you ask back for the problem statement, put these in the same round; otherwise put them above the candidates in step 3 and go on without waiting for the answers.

  2. Before generating candidates in step 3, look for what already exists but is not used for this problem: in a codebase, machinery that sits disconnected; elsewhere, resources, programs, people or partnerships already in reach.

    a. Search the material yourself or use the unknowns:unknowns-scout agent.

    b. When you spawn the scout, the packet overrides the scout's default output (its finding table and improved prompt draft). The packet gives the scout the problem statement, the constraints and where to search, and asks it to return an inventory, one row per existing mechanism that bears on the problem, with the columns where it is / what it does today / what is missing to put it to use / evidence / status (confirmed / inferred); the scout is done when its procedure is done and every mechanism it found has a row.

    c. When the scout returns, put its Questions to answer, Information still needed and Needs checking lists above the candidates in step 3, for the user to answer or check.

    Without a tool, do the step in the conversation and say what could not be checked.

    The cheapest candidates are usually connecting what exists, not building.

  3. Generate ~10 candidate interventions with judgment deferred, spread across the time axis (this afternoon / short-term (up to a month) / mid-term (1–3 months) / quarter-long bets (longer)) and across approaches, the unlikely ones included. Give each only:

    • name
    • one-line description: what it would change
    • Already exists — <where>, when the candidate builds on something step 2 found

    Ranking, pruning, sizing and any recommendation wait for the user's picks (steps 4–5). Judging while generating keeps the first safe ideas and stops the list before the unexpected ones (Osborn: defer judgment, go for quantity).

  4. End the reply by asking which candidate numbers resonate; those are the picks. The picks show the user's priorities, which you do not know yet; a recommendation added after the list makes the pick for them.

  5. Once the user names the picks, attach these fields to each picked candidate:

    • expected impact: what improves, by how much
    • effort size: S/M/L/XL
    • key risks
    • how to measure success: the sign that it moved the stated problem, not only that it was done

    The picked numbers assemble into a copyable "proceed with these, in this order"; a missing pick is undecided.

Output

Layout: numbered candidates on a time axis (this afternoon → quarter-long bet).

Related

After step 5, suggest the next skill by what is still open:

Still open Skill
The form is in question prototypes
Decisions only the user can make interview
The plan for the work tweakable-plan

04 同仓库更多 Skills