agentleFS
Shoko-officialShoko-official/Claude-Science-System-Prompts/skills/conducting-scientific-research/SKILL.md

conducting-scientific-research

Skill10 starsChanged 3 months agoApache-2.0View on GitHub

Conduct rigorous, reproducible multi-step scientific work with literature, databases, local files, Python, R, shell, artifacts, reviewers, and approved compute. Use for evidence synthesis, data or statistical analysis, machine learning, simulation, study design, scientific figures or manuscripts, reproduction audits, and database curation. Do not invoke for a simple timeless science fact that needs no tools or project workflow. Kaggle competitions use the separate kaggle-competition skill.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 8 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill when the task is scientific work rather than a single factual explanation.

    line 8
  2. Distinctive:

    Read the project instructions and inspect the referenced files, artifacts, prior sessions, environments, connectors, compute, and reviewer …

    line 12
  3. Distinctive:

    Read the relevant references below before the first substantive action.

    line 14
  4. Distinctive:

    Read only the references that affect the active work.

    line 26
  5. Distinctive:

    Never report execution, retrieval, validation, review, or saving unless the record proves it.

    line 30
  6. Distinctive:

    Keep source claims, direct observations, computed values, inferences, and hypotheses distinct.

    line 31
  7. Distinctive:

    Do not rely on hidden kernel state for a durable result.

    line 33
  8. Distinctive:

    Use the least permission necessary.

    line 35

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “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

Conducting Scientific Research

Use this skill when the task is scientific work rather than a single factual explanation. Adapt the procedure to the request; do not force every task through every reference.

Start

  1. Read the project instructions and inspect the referenced files, artifacts, prior sessions, environments, connectors, compute, and reviewer state.
  2. Define the requested result and the smallest evidence or execution path that can support it.
  3. Read the relevant references below before the first substantive action.
  4. Identify the validation gate, durable artifacts, and any new permission or external-action boundary.
  5. Execute, validate, save the record, request review when material, address findings, and report the result.

Reference routing

  • Scientific questions, study design, exploration versus confirmation, and manuscripts: references/scientific-work.md
  • Literature search, citation checking, evidence tables, and database retrieval: references/literature-and-retrieval.md
  • Data audit, statistics, causal inference, machine learning, and figures: references/data-statistics-ml.md
  • Environments, local and remote compute, artifacts, provenance, and review: references/compute-artifacts-review.md
  • Reusable project records and templates: references/templates.md

Read only the references that affect the active work. Keep reference loading one level deep.

Required behavior

  • Never report execution, retrieval, validation, review, or saving unless the record proves it.
  • Keep source claims, direct observations, computed values, inferences, and hypotheses distinct.
  • Preserve raw inputs and material identity: units, builds, versions, identifiers, filters, joins, exclusions, and query dates.
  • Do not rely on hidden kernel state for a durable result; save code and rerun from declared inputs when practical.
  • Validate fragile retrievals, joins, models, figures, and artifacts with an independent check.
  • Use the least permission necessary. Do not expose credentials or perform an external action without the required approval.

Default loop

Inspect state → establish objective and evidence → execute → validate
→ save artifacts and provenance → review → correct → report

For a simple task, several stages may collapse into one. For a material task, do not omit validation or the durable record merely to finish faster.

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.