agentleFS
ralfyishereralfyishere/rules-with-receipts/.claude/skills/divergent-ideation/SKILL.md

Divergent Ideation

Skill2 starsChanged 4 months agoMITView on GitHub

Generate different options before choosing - for brainstorms, naming, design alternatives, and any open problem where the first workable idea becomes the only idea. Activate on "brainstorm", "give me ideas/options/names", "what could we build", "how else could we do this", first attempts on open-ended problems, and when your "options" all share one core mechanism. PROACTIVELY on open/improve/turnaround mandates ("improve this", "what would you do", "print money", "do what you think") - generate unplanted angles instead of waiting for direction. Not for problems with a known correct answer.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    When the user has chosen and wants execution: re-opening the option space uninvited is scope-fence territory.

    line 23
  2. Very specific:

    Subtract: remove the feature/step/component instead of adding one.

    line 37
  3. Distinctive:

    Safe-list syndrome: every option is defensible and none is interesting — usually means step 4 was skipped and no assumption was inverted.

    line 56
  4. Distinctive:

    Ask: "Ideas for reducing our app's support ticket volume." Obvious first three: better docs, chatbot, FAQ.

    line 63

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Divergent Ideation

Purpose

The default failure of idea generation isn't bad ideas — it's narrow ones: the first workable concept arrives, and everything after is that concept wearing different hats. Real option value comes from ideas with different core mechanisms. This skill separates divergence (generate wide, judgment off) from convergence (filter hard, judgment on), because doing both at once does both badly.

When to use this skill

  • Explicit generation asks: brainstorms, naming, campaign concepts, feature ideas, alternative designs.
  • The first attempt at any open-ended problem — before committing to an approach that arrived unexamined (plan-gate's risk section deserves real alternatives).
  • Mid-task, when stuck: circling one approach that isn't working is a divergence trigger, not a push-harder trigger.
  • Open / improve / turnaround mandates ("improve this", "what would you do", "do what you think"): generate unplanted angles proactively instead of waiting to be handed a direction — this is where the first idea most silently becomes the only idea. (Distinct from open-mandate, which chooses the direction; this skill first widens the option set the choice is made from.)

When NOT to use this skill

  • Problems with a verifiable correct answer — go verify, don't ideate.
  • Established-convention decisions (code formatting, standard patterns) where novelty is a cost, not a value.
  • When the user has chosen and wants execution: re-opening the option space uninvited is scope-fence territory.

Operating procedure

1 — Frame the actual problem (one line, from intent-clarity): what the ideas must accomplish and the 2–3 real constraints. Under-constrained ideation produces volume without relevance; over-constrained reproduces the status quo.

2 — Diverge with judgment OFF. Generate 8–15 raw options. The count is functional, not decorative: the first ~3 ideas are the obvious ones everyone has; the different-mechanism ideas live past them. No evaluating, no "but that wouldn't work" — evaluation mid-generation kills exactly the unusual candidates this phase exists to surface.

3 — Force variety with axes (when generation converges, switch axes):

  • Invert an assumption: what if the constraint everyone accepts were false?
  • Change the actor: solved by the user themselves? by automation? by a partner? by nobody (eliminate the need)?
  • Change the resource frame: 10× the budget; zero budget; 10× the timeline; ship tomorrow.
  • Steal across domains: how do games / logistics / restaurants / biology solve the analogous problem?
  • Combine: pair two mediocre options and check the hybrid.
  • Subtract: remove the feature/step/component instead of adding one.

4 — Include deliberately "wrong" candidates. At least two ideas that are infeasible as stated — too expensive, too weird, against policy. Infeasible ideas are frequently a feasible idea plus one removable defect, and they stretch the space the feasible ideas get drawn from.

5 — Converge with judgment ON. Now switch modes: cluster near-duplicates (count each mechanism once), name the selection criteria, and score honestly — hand real decisions to structured-reasoning (matrix or tradeoff). Select 2–3, each with a one-line why.

6 — Present the shortlist AND the longlist. Lead with your 2–3 picks and reasoning; attach the full list compactly below. The user's context differs from yours — their winner is regularly one you'd have culled, and showing discards costs three lines.

Quality bar

  • The shortlisted options have different core mechanisms — the test: does each one fail for different reasons? Options that would all fail the same way are one option.
  • Generation genuinely preceded evaluation (detectable: the longlist contains ideas the shortlist reasoning argues against).
  • At least two axes from step 3 were used — the list isn't fifteen synonyms of the first idea.
  • Constraints from step 1 were honored by the shortlist (the longlist is allowed to break them; that's what it's for).

Common failure modes

  • First-idea anchoring: idea #1 arrives, ideas #2–10 are its variations. Check mechanisms, not surface descriptions.
  • Premature convergence: evaluating each idea as it's generated, killing the weird ones in utero. The phases exist because the mindsets are incompatible.
  • Safe-list syndrome: every option is defensible and none is interesting — usually means step 4 was skipped and no assumption was inverted.
  • Volume theater: twenty options, three mechanisms — padding a narrow list with rephrasings to hit a count. The count serves diversity; diversity is the target.
  • Hidden discards: presenting only the winners. The discarded list is cheap to include and regularly contains the user's actual pick.
  • Ideating past the decision: the user said "option B, go" and the response reopens the space. Divergence has a season; respect its end.

Example

Ask: "Ideas for reducing our app's support ticket volume." Obvious first three: better docs, chatbot, FAQ. Axis switches — change the actor: let power users answer (community forum with rep points); invert the assumption ("tickets are inevitable"): instrument the top-5 ticket-generating screens and fix the confusion at the source; subtract: remove the feature that generates 20% of tickets and serves 2% of users; cross-domain (restaurants): a "kitchen open" livestream — public real-time status page killing "is it down?" tickets; "wrong" idea: charge for support (infeasible as stated → kernel: free tier gets community support, paid gets human). Shortlist: instrument-and-fix (attacks cause), status page (cheap, kills a whole category), community tier (scales). Different mechanisms: one reduces confusion, one reduces uncertainty, one redistributes load. Longlist attached.

Works with sibling skills

  • intent-clarity frames step 1; structured-reasoning owns step 5's selection — this skill builds the option set that makes selection meaningful.
  • product-thinking supplies the value lens when the ideas are product ideas; failure-mode-awareness stress-tests the shortlist afterward.
  • The convergent skills (adversarial-verify, verification-discipline) stay out of step 2 by design — they re-enter at step 5.

Provenance and maintenance

Added 2026-07 in the expansion pass: every core skill is convergent (narrowing, verifying, filtering); generation had no owner, and first-idea anchoring is the observable result. Techniques are standard creative-process craft. Re-verify by mechanism-counting recent brainstorm outputs: shortlists whose options all share one mechanism mean steps 3–4 aren't firing.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.