From paper-reading-skills
Transforms Research Opportunity Candidates from a Field Map into decision-ready research questions with evidence sufficiency checks and structured cards.
How this skill is triggered — by the user, by Claude, or both
Slash command
/paper-reading-skills:02-research-questionThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use this skill to run a Research Question Workflow: select a source Field Map Workspace, confirm the user's Writing Intent, turn Research Opportunity Candidates into Candidate Angles, check whether the evidence is sufficient, collect Supporting Evidence when needed, and produce Research Question Cards.
Use this skill to run a Research Question Workflow: select a source Field Map Workspace, confirm the user's Writing Intent, turn Research Opportunity Candidates into Candidate Angles, check whether the evidence is sufficient, collect Supporting Evidence when needed, and produce Research Question Cards.
This skill does not start from generic brainstorming. It starts from one Field Map Workspace and carries its Research Opportunity Candidates forward into decision-ready research questions.
Set {workspace-root} before creating, scanning, or updating artifacts:
{workspace-root} to workspace (the repo-local workspace/ directory).workspace/ layer.{workspace-root} from that path and keep related artifacts under the same root.{workspace-root}/research-questions/{field-slug}/.source_field_map.md.writing_intent.md.research_opportunities.md, research_clusters.md, and relevant Paper Position Records.candidate_angles.md.evidence/.cards/.research_question_cards.md.keep Research Question Cards, or one unusually strong keep card with a recorded reason for not keeping more.If a Research Question Workspace already exists, read its current artifacts before continuing. Preserve user edits and update the existing files instead of overwriting them blindly.
The Research Question Workflow must be derived from exactly one Field Map Workspace.
If the user provides a slug or path, inspect that workspace. If the user does not provide one, scan {workspace-root}/field-maps/ for complete workspaces.
A Field Map Workspace is complete enough for this workflow only when it contains:
field_boundary.mdresearch_clusters.mdresearch_opportunities.mdseed_papers/ or all_papers/Routing rules:
paper-reading-field-map; do not invent Research Opportunity Candidates.research_opportunities.md is missing, the user needs the Field Map Workflow's opportunity step before Research Question Cards can be produced.Record the source workspace and completeness check in source_field_map.md using references/source-field-map-template.md.
Create durable artifacts at:
{workspace-root}/research-questions/{field-slug}/
Use this structure:
source_field_map.mdwriting_intent.mdcandidate_angles.mdevidence/cards/research_question_cards.mdUse references/workspace-structure.md as the structure guide.
Use Guided Choice whenever the workflow needs the user to decide between plausible directions.
Do not present one recommendation and ask the user to confirm it unless the user has already stated a clear preference. Instead, present:
Good Guided Choice prompts are comparative:
When presenting multiple options, explain what each option is best for and what it sacrifices. If the user chooses one branch, continue narrowing from that branch instead of reopening every option.
Do not recommend Candidate Angles until the user confirms the Writing Intent.
Start by offering a Guided Choice over plausible Writing Intent profiles from the user's stated goal. Do not ask the user to confirm a single proposed profile.
The user should choose:
Default goal vocabulary:
Useful practical constraints include available time, number of collaborators, ability to run user studies, ability to release a benchmark, target research community, target venue type, compute budget, data access, and preferred contribution type.
If the user is unsure, use binary narrowing questions before writing the final Writing Intent:
Write the confirmed Writing Intent to writing_intent.md using references/writing-intent-template.md.
A Candidate Angle is a narrower research cut derived from a Research Opportunity Candidate. Each Candidate Angle must include a Contribution Type.
Allowed Contribution Types:
For each Research Opportunity Candidate, derive 2-5 Candidate Angles when possible. Each Candidate Angle should include:
Write Candidate Angles to candidate_angles.md using references/candidate-angles-template.md.
After writing Candidate Angles, present a Guided Choice set rather than a single winner. The set should include several recommended angles with different strengths, such as:
Ask the user to select 1-3 angles or answer a binary narrowing question. Do not create Research Question Cards before the user selects angles.
Use Writing Intent to recommend Candidate Angles:
Benchmark / Evaluation Protocol and Empirical Study / Measurement.Theory / Conceptual Model angles with strong evidence that the field lacks a stable framing.System / Tooling angles with a clear architecture, workload, evaluation environment, and integration story.If Writing Intent goals conflict, explain the tradeoff instead of forcing one score to hide it.
When multiple Candidate Angles fit the Writing Intent, keep the top several options visible. Recommend a short list and say which one you would choose under each plausible user priority.
Run an Evidence Sufficiency Check after the user selects Candidate Angles and before generating Research Question Cards.
A Candidate Angle is minimally sufficient when it has:
Evidence roles include:
Decision outcomes:
sufficient: create or update a Research Question Card.needs more evidence: collect Supporting Evidence before creating a keep card.defer: do not create a keep card unless the user explicitly wants a speculative record; mark unsupported claims as evidence needs.If evidence is insufficient but the Candidate Angle is promising, collect Supporting Evidence under:
{workspace-root}/research-questions/{field-slug}/evidence/
Use references/supporting-evidence-template.md.
Supporting Evidence should record search questions, keyword queries, discovery sources, new papers, evidence roles, useful evidence points, remaining gaps, and how the evidence changes the Candidate Angle.
If network access is unavailable, do not pretend the evidence collection is complete. Tell the user what could not be searched and ask for paper links or permission to continue when network access is available.
If a newly found paper becomes central to the field map itself, recommend adding it to the Field Map Workspace later, but do not block the Research Question Workflow unless the Research Question Card would otherwise rely on uncited claims.
Each Research Question Card must cite evidence. Do not make core claims without links to Field Map paper records or Supporting Evidence files.
Use references/research-question-card-template.md for each card in:
{workspace-root}/research-questions/{field-slug}/cards/{question-slug}.md
Each card must include:
keep, needs more evidence, or deferUpdate research_question_cards.md using references/research-question-cards-summary-template.md.
Do not skip these gates:
keep unless its core claims cite evidence.Use the reference templates in this directory:
references/workspace-structure.mdreferences/source-field-map-template.mdreferences/writing-intent-template.mdreferences/candidate-angles-template.mdreferences/supporting-evidence-template.mdreferences/research-question-card-template.mdreferences/research-question-cards-summary-template.mdStop when the user has:
If only one Research Question Card is strong enough to keep, stop only after recording why the workflow did not keep more.
npx claudepluginhub snake-fan/paper-reading-skillsCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.