Idea Finder
Read ../researchstack/references/workflow.md, ../researchstack/references/venues.md, ../researchstack/references/conference-story-patterns.md, ../researchstack/references/novelty-boundary.md, ../researchstack/references/evidence-contracts.md, and ../researchstack/references/memory.md.
First decide whether topic choice is actually open.
- If the user wants to preserve a topic, hand off to
researchstack-method-synthesis and do not return replacement topics.
- If the user is planning several papers under one direction, hand off to
researchstack-program-map.
- Continue here only when problem discovery is wanted.
Flow
- Establish area, venue family, resource constraints, available artifacts, and excluded directions.
- Search recent papers, official systems, and mature open-source implementations.
- Map concrete failures, unexplained phenomena, deployment mismatches, and missing capabilities.
- Generate a small candidate set across different paper types; do not vary only the application label.
- For each candidate, identify the closest topic and method competitors.
- Run
researchstack-idea-review and the one-paper envelope.
- Surface only candidates with an honest contribution and feasible evidence path.
Reject trend labels without a research question, topic novelty backed only by terminology, and directions whose method is ordinary engineering. Do not reject a crowded topic solely because it is crowded; route it to method synthesis when the problem remains important.
For each survivor use assets/topic-card-template.md and include:
- one-paper thesis and paper type,
- why the problem matters now,
- topic state and method state,
- closest topic and method competitors,
- irreducible contribution,
- one or two falsifiable claims,
- minimum evidence path,
- top reviewer objection,
- explicit out-of-scope work,
- idea-review verdict.
Use assets/rejected-directions-template.md for rejected candidates. Distinguish problem weak, method weak, evidence infeasible, and already done rather than collapsing them into one novelty label.
1---2name: researchstack-idea-finder-23description: Problem-discovery skill for systems, networking, and AI research. Use when the user has a broad area and is willing to choose among paper-sized problems. Do not use it to force a topic change when the user has fixed the topic and needs a new method; use method-synthesis instead.4---56<!-- AUTO-GENERATED for codex from researchstack v0.3.0. Edit source SKILL.md files, then regenerate from the source checkout. -->78# Idea Finder910Read [../researchstack/references/workflow.md](../researchstack/references/workflow.md), [../researchstack/references/venues.md](../researchstack/references/venues.md), [../researchstack/references/conference-story-patterns.md](../researchstack/references/conference-story-patterns.md), [../researchstack/references/novelty-boundary.md](../researchstack/references/novelty-boundary.md), [../researchstack/references/evidence-contracts.md](../researchstack/references/evidence-contracts.md), and [../researchstack/references/memory.md](../researchstack/references/memory.md).1112First decide whether topic choice is actually open.1314- If the user wants to preserve a topic, hand off to `researchstack-method-synthesis` and do not return replacement topics.15- If the user is planning several papers under one direction, hand off to `researchstack-program-map`.16- Continue here only when problem discovery is wanted.1718## Flow19201. Establish area, venue family, resource constraints, available artifacts, and excluded directions.212. Search recent papers, official systems, and mature open-source implementations.223. Map concrete failures, unexplained phenomena, deployment mismatches, and missing capabilities.234. Generate a small candidate set across different paper types; do not vary only the application label.245. For each candidate, identify the closest topic and method competitors.256. Run `researchstack-idea-review` and the one-paper envelope.267. Surface only candidates with an honest contribution and feasible evidence path.2728Reject trend labels without a research question, topic novelty backed only by terminology, and directions whose method is ordinary engineering. Do not reject a crowded topic solely because it is crowded; route it to method synthesis when the problem remains important.2930For each survivor use [assets/topic-card-template.md](assets/topic-card-template.md) and include:3132- one-paper thesis and paper type,33- why the problem matters now,34- topic state and method state,35- closest topic and method competitors,36- irreducible contribution,37- one or two falsifiable claims,38- minimum evidence path,39- top reviewer objection,40- explicit out-of-scope work,41- idea-review verdict.4243Use [assets/rejected-directions-template.md](assets/rejected-directions-template.md) for rejected candidates. Distinguish `problem weak`, `method weak`, `evidence infeasible`, and `already done` rather than collapsing them into one novelty label.
lqf0624