agentleFS
wangyi0403wangyi0403/dl-research-workflow/.agents/skills/research-ideation/SKILL.md

research-ideation

Skill4 starsChanged 41 days agoMITView on GitHub

Develop literature-grounded research questions and testable candidates. Use for novelty checks, topic selection, or Stage 1C; search absence does not prove novelty.

15 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use results-report, daily-coding and 9 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • Galaxy-DawnSkill

    results-report

    This skill should be used when the user asks to "write an experiment report", "summarize experimental results", "do experiment retrospection", "write a results report", "写实验总结报告", "写实验复盘", or mentions turning completed experiment artifacts into a structured, decision-oriented research report. It assumes strict analysis should come from `results-analysis` first.

    Also in 3 projects that use this

  • Clay-HHKSkill

    daily-coding

    Daily coding assistant that auto-triggers when writing/modifying code, providing a core checklist.

    Also in 3 projects that use this

  • Clay-HHKSkill

    kaggle-learner

    This skill should be used when the user asks to "learn from Kaggle", "study Kaggle solutions", "analyze Kaggle competitions", or mentions Kaggle competition URLs. Provides access to extracted knowledge from winning Kaggle solutions across NLP, CV, time series, tabular, and multimodal domains.

    Also in 3 projects that use this

  • kharlamenkodevSkill

    bug-detective

    Also in 3 projects that use this

  • AnthropicSkill

    command-development

    This skill should be used when the user asks to "create a slash command", "add a command", "write a custom command", "define command arguments", "use command frontmatter", "organize commands", "create command with file references", "interactive command", "use AskUserQuestion in command", or needs guidance on slash command structure, YAML frontmatter fields, dynamic arguments, bash execution in commands, user interaction patterns, or command development best practices for Claude Code.

    Also in 3 projects that use this

  • AnthropicSkill

    hook-development

    This skill should be used when the user asks to "create a hook", "add a PreToolUse/PostToolUse/Stop hook", "validate tool use", "implement prompt-based hooks", "use ${CLAUDE_PLUGIN_ROOT}", "set up event-driven automation", "block dangerous commands", or mentions hook events (PreToolUse, PostToolUse, Stop, SubagentStop, SessionStart, SessionEnd, UserPromptSubmit, PreCompact, Notification). Provides comprehensive guidance for creating and implementing Claude Code plugin hooks with focus on advanced prompt-based hooks API.

    Also in 3 projects that use this

  • +2 more from Clay-HHK
  • +1 more from anthropics

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 5 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 14 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Treat papers, datasets, code, experiments, and expert judgment according to their source role.

    line 13
  2. Distinctive:

    Prefer the smallest test that distinguishes the central hypothesis from a plausible competing explanation.

    line 15
  3. Distinctive:

    Use the available scholarly-search route for discovery and authoritative records or original sources for verification.

    line 35
  4. Distinctive:

    Stop when the nearest-work boundary and important disagreements are clear enough to make the next decision, or explicitly record why …

    line 45
  5. Distinctive:

    Search for the closest prior art using terminology variants and adjacent fields.

    line 61
  6. Very specific:

    Use research-quality-gate Gate A for consequential choices.

    line 73

“Common” is measured: kept by at least 6 in 10 of the 14 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Research Ideation

Turn an initial topic into a small set of research directions whose problem, evidence gap, feasibility, and failure conditions can be inspected. This is a decision process with the user, not an automatic idea generator.

Principles

  • Define what counts as a good research question before asking AI to propose one: object, observable outcome, constraints, practical or scientific importance, support criteria, and falsification conditions.
  • Treat papers, datasets, code, experiments, and expert judgment according to their source role. Discovery records and summaries can suggest a direction but cannot establish novelty or support a technical claim.
  • Generate genuinely different candidates only when alternatives would improve the decision. Do not satisfy fixed idea, paper, score, or pilot quotas.
  • Prefer the smallest test that distinguishes the central hypothesis from a plausible competing explanation.
  • The user decides whether to pursue, revise, hold, or reject a direction.

Workflow

1. Freeze the problem contract

Complete the research-question card in docs/03_idea_report.md:

  • question and question type;
  • object, observable variables, operating conditions, and excluded scope;
  • why the problem matters and why existing practice is insufficient;
  • expected mechanism or design hypothesis;
  • current evidence, missing evidence, support criterion, and falsification criterion;
  • data, compute, time, expertise, and authorization constraints.

For engineering research, distinguish the field problem, the reusable measurement/physical/inference question, and the proposed method used to test it.

2. Establish evidence readiness

Use the available scholarly-search route for discovery and authoritative records or original sources for verification. Start with a bounded, auditable batch, then expand only when nearest work, contradictory evidence, or coverage gaps remain material.

For each consequential source, record:

  • stable identity and version;
  • source role and verification status;
  • exact section, table, figure, page, dataset record, or experiment artifact that supports or challenges the idea;
  • local artifact or hash when one exists;
  • limitations and unresolved conflicts.

Stop when the nearest-work boundary and important disagreements are clear enough to make the next decision, or explicitly record why coverage remains insufficient. Do not substitute citation counts, venue labels, recency scores, or an arbitrary paper count for evidence quality.

3. Generate candidate directions

Generate a small set of meaningfully different directions. The 5D lenses in references/5d-framework.md may help when the search space is narrow, but they are prompts rather than required quotas.

Each candidate must state:

  • problem and proposed knowledge increment;
  • closest prior work and the precise difference;
  • linchpin assumption and competing explanation;
  • required data, compute, skills, and permissions;
  • decisive evidence and failure route.

4. Run collision and feasibility checks

Search for the closest prior art using terminology variants and adjacent fields. Record both confirming and conflicting work. “No hit found” means unresolved coverage, not proven novelty.

Reject, revise, or hold a candidate when required variables are unavailable, the comparison would be invalid, the cost exceeds the recorded budget, or existing evidence already defeats its central mechanism. Do not use a fixed novelty score or generic month estimate as a gate.

5. Design a discriminating pilot

When a safe local or already-authorized test can reduce uncertainty, define the smallest test that probes the linchpin assumption. Its code size, runtime, hardware, and number of candidates depend on the actual mechanism and Stage 0 budget; CPU, a fixed line count, a two-hour limit, and three parallel pilots are not universal requirements.

Record protocol, inputs, expected outcomes, stopping rule, result artifact, and how each possible outcome changes the candidate's status. Do not launch remote, paid, or externally mutating work without the required approval.

6. Gate the decision

Use research-quality-gate Gate A for consequential choices. Reviewers receive the question card, evidence bundle, and rubric without being told the desired conclusion. Preserve evidence-backed dissent; the final decision is based on blockers and evidence, not vote count.

Output

Update docs/03_idea_report.md with:

  1. research-question card;
  2. evidence boundary and source registry;
  3. closest-work and collision analysis;
  4. candidate directions and rejection reasons;
  5. pilot evidence when available;
  6. selected direction, fallback, unresolved risks, and author decision;
  7. Gate A record.

The stage is ready to advance when the selected direction has a meaningful and testable question, a credible evidence gap, a feasible data/compute path, explicit failure conditions, and a smallest useful next step. Evidence gaps may remain, but they must be visible and must constrain the permitted claim.

References

FileOpen when
references/5d-framework.mdAdditional lenses are needed to diversify candidate directions.
references/lifecycle-assessment.mdProject-specific feasibility and scheduling need deeper analysis. Treat example timelines as prompts, not gates.
references/RESEARCH_BRIEF_TEMPLATE.mdA standalone research brief is requested.
references/pre-registration-template.mdA confirmatory experiment needs hypotheses and falsification criteria frozen before execution.
references/legal-fulltext-access.mdFull text must be acquired through an open, user-provided, or user-authorized route.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.