Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →brycewang-stanford/workflows:ideate
Divergent research ideation — generate many candidate directions, then adversarially filter to the strongest
Passed 9 of 22 test cases with the skill, and 4 of 22 without it.
and 5 more in the full report
6 fixed3 passed either way12 still failing1 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
We are looking for fresh academic project ideas on mitigating hallucination in retrieval-augmented generation systems for legal documents. Please generate a deep pool of potential research candidates across different computational and data angles before narrowing them down. Put the candidate set through a rigorous critique to extract the strongest proposals, and write up a structured ideation report with frontmatter metadata, survivor rankings, and clear reasons for rejected concepts.
| Checked | Without skill | With skill |
|---|---|---|
| Generates between 15 and 20 initial candidate research directions. | Missed | Passed |
| Includes candidates_generated and survivors fields in the YAML frontmatter. | Missed | Passed |
| Provides a single-line rejection reason for each non-surviving candidate. | Passed | Passed |
The instructions your model reads when the skill loads.
Divergent exploration before convergent brainstorming. Generate many candidates, then filter ruthlessly.
Read $ARGUMENTS. If the user provides a specific research question, ideate around it. If they provide a broad topic, explore broadly.
Generate 15-20 candidate research directions. Use these research-adapted ideation frames:
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.
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:
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.
Write the ideation document to docs/ideation/ with YAML frontmatter:
test cases passed, vs 4 / 22 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.