Researchstack Next Step

Next Step

Read ../references/workflow.md and ../references/memory.md.

If the project is ongoing, inspect relevant project memory first via the researchstack-learn workflow. Pull in only high-signal context:

• current thesis,
• venue and deadline,
• known reviewer risks,
• experiment status,
• artifact or reproducibility constraints,
• writing or framing preferences,
• rejected directions.

This skill is a decision layer, not an execution skill.

The job is to answer:

Given the current research state, what should the user do next?

Required output:

1. Current stage.
2. Blocking mismatch.
3. Recommended next skill, usually exactly one and at most two.
4. Why this step beats the alternatives.
5. What not to do yet.
6. A short follow-up chain after the recommended step.

Use assets/next-step-decision-template.md.

Stage labels:

• research-program planning,
• idea formation,
• method formation,
• literature positioning,
• evidence building,
• experiment operations,
• artifact hardening,
• writing,
• layout and figure polish,
• submission triage,
• review and rebuttal.

Common blocking mismatches:

• a durable topic exists but paper candidates overlap,
• the topic is fixed but the proposed method collapses into prior work,
• topic novelty is being confused with method novelty,
• thesis exists but novelty is untested,
• thesis exists but related-work boundary is unclear,
• claims are broader than current evidence,
• results exist but artifact provenance is weak,
• figures exist but claim-to-evidence mapping is unclear,
• writing started before experiment design is locked,
• deadline pressure means scope should shrink rather than expand,
• reviewer risk is known but not answered by an experiment.

Decision rules:

• If the user needs several distinct papers under one topic, recommend researchstack-program-map.
• If the thesis is vague, recommend researchstack-lab-intake or researchstack-idea-refine.
• If the user only has an area, recommend researchstack-idea-finder.
• If the topic is fixed but the method is weak or already has close precedent, recommend researchstack-method-synthesis.
• If novelty is untested, recommend researchstack-idea-review.
• If the closest competitors are unclear, recommend researchstack-literature-map.
• If claims and evaluation are not aligned, recommend researchstack-experiment-design.
• If experiments are chaotic or results are hard to promote, recommend researchstack-experiment-ops.
• If numbers or figures cannot be traced, recommend researchstack-artifact-audit.
• If implementation could invalidate conclusions, recommend researchstack-code-review.
• If evidence is stable and the thesis is locked, recommend researchstack-paper-write.
• If writing exists but figures/tables/captions are weak, recommend researchstack-figure-studio or researchstack-paper-layout.
• If the user is near a deadline, recommend researchstack-submission-gate.
• If a draft is complete enough to attack, recommend researchstack-peer-review.
• If reviews exist, recommend researchstack-rebuttal-coach.

Be decisive. Do not list the whole skill catalog.

If there is not enough information to decide, ask for only the missing state needed to choose the next skill:

• current thesis,
• strongest current evidence,
• closest baseline or related paper,
• whether the topic must remain fixed,
• venue or deadline,
• what artifact/code/result exists now.

When a durable decision emerges, suggest recording it with researchstack-learn.