Skip to content

brycewang-stanford/Auto-Empirical-Research-Skills / skills/11-James-Traina-compound-science/skills/workflows-ideate

workflows:ideate

Verified

Divergent research ideation — generate many candidate directions, then adversarially filter to the strongest

Research Ideation

Divergent exploration before convergent brainstorming. Generate many candidates, then filter ruthlessly.

Phase 0: Scope the Ideation

Read $ARGUMENTS. If the user provides a specific research question, ideate around it. If they provide a broad topic, explore broadly.

Phase 1: Generate Candidates (Divergent)

Generate 15-20 candidate research directions. Use these research-adapted ideation frames:

  1. Identification weakness — What existing results have weak identification? What new variation could fix it?
  2. Computational bottleneck — What problems are infeasible with current methods but tractable with new estimators or hardware?
  3. Data limitation workaround — What would become possible with data that is now available but underexploited?
  4. Alternative estimator class — What if the standard approach (e.g., linear IV) were replaced with a different class (e.g., ML, structural, Bayesian)?
  5. Relaxed assumption — What results depend on assumptions that could be relaxed? What happens when you relax them?
  6. Literature gap — What do practitioners need that academics haven't provided? What do adjacent fields know that this field doesn't?

Dispatch methods-explorer and literature-scout agents in parallel to ground the ideation in real methods and recent papers.

Iron rule: Generate the full candidate list before critiquing any idea. Push past the first few obvious directions.

Phase 2: Adversarial Filter (Convergent)

Entry condition: Phase 1 produced at least 15 candidate directions (the iron rule). Exit condition: 5-7 survivors identified, all rejected candidates have one-line rejection reasons.

For each candidate, evaluate:

  • Feasibility (0-100): Can this be done with available data, methods, and time?
  • Contribution (0-100): Would a top journal care about this result?
  • Identification (0-100): Is there a credible identification strategy?

Dispatch identification-critic to attack the top 10 candidates. Only candidates surviving adversarial scrutiny advance.

Target: 5-7 survivors with explicit rejection reasons for all others.

Phase 3: Output

Write the ideation document to docs/ideation/ with YAML frontmatter:

---
status: complete
date: YYYY-MM-DD
topic: <descriptive topic>
candidates_generated: <N>
survivors: <N>
---

Content:

  • Surviving candidates ranked by (Contribution x Identification x Feasibility)
  • For each survivor: 2-3 sentence description, confidence score, key risk
  • Rejected candidates with one-line rejection reason
  • Recommended next step: /workflows:brainstorm on the top 1-2 candidates

Handoff

End with: "Ideation complete. Run /workflows:brainstorm [top candidate] to develop requirements for the strongest direction."

add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution.Stars: 115kUpdated Claude CodeCodexVerifiedInstall
add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation.Stars: 115kUpdated Claude CodeCodexVerifiedInstall
diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope.Stars: 115kUpdated Claude CodeCodexVerifiedInstall
fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks.Stars: 115kUpdated Claude CodeCodexVerifiedInstall