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

research-refine

Skill4 starsChanged 41 days agoMITView on GitHub

Refine an accepted, evidence-backed idea into a minimal method, mechanism hypothesis, failure modes, and claim-driven validation plan.

8 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 8 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read AGENTS.md, docs/03_idea_report.md, data analysis, verified literature/evidence records, and any approved constraints.

    line 12
  2. Very specific:

    Run a focused review: Use the project's local review procedure or $research-quality-gate to test problem fidelity, mechanism specificity, …

    line 22
  3. Distinctive:

    Do not turn an unverified observation into a mechanism, law, novelty claim, or physical contribution.

    line 26
  4. Distinctive:

    Do not solve a different, easier problem merely because it makes a cleaner model demo.

    line 27
  5. Distinctive:

    Prefer a small mechanism that can be disproved over a broad system with many loosely connected innovations.

    line 28
  6. Distinctive:

    Do not promise expensive training, external review, upload, or publication without the required authorization.

    line 29

“Common” is measured: kept by at least 6 in 10 of the 8 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 Refine

Refine an approved research direction without allowing it to drift from the problem it is meant to solve. The output is a falsifiable method-and-validation plan, not a feature list.

Inputs and output

Read AGENTS.md, docs/03_idea_report.md, data analysis, verified literature/evidence records, and any approved constraints. Update docs/04_methodology.md; record material decisions in docs/03_idea_report.md or the project decision log.

Workflow

  1. Freeze the problem anchor. State the real problem, affected setting, measurable consequence, existing-solution bottleneck, and the exact research question. Mark assumptions and evidence gaps.
  2. State the mechanism hypothesis. Identify observable signals, nuisance factors, representation or inference failure, and a falsifiable prediction. For engineering work, distinguish a physical/measurement contribution from a useful empirical hypothesis; call it physical only when its mechanism and test are concrete.
  3. Derive one candidate contribution. Express a transferable design principle or knowledge claim before naming the method. Explain how it differs from closest work using verified evidence.
  4. Design the smallest sufficient method. Define inputs, outputs, modules, interfaces, training/inference path, and why each part tests or enables the hypothesis. Remove any module with no diagnostic, claim, or control purpose.
  5. Specify failure modes. List expected non-identifiability, distribution shift, weak-signal, label, data-coverage, and resource failure modes, including how each would be detected.
  6. Build the validation chain. Map every Claim ID to the required data, baseline/control, metric, ablation or intervention, uncertainty treatment, and stopping condition. Engineering work must separately test operational value, mechanism/design-principle evidence, and any transfer claim.
  7. Run a focused review. Use the project's local review procedure or $research-quality-gate to test problem fidelity, mechanism specificity, parsimony, evidence sufficiency, feasibility, and drift. Record findings in the active Stage artifact; do not assume an unavailable sub-agent or external model.

Non-negotiable constraints

  • Do not turn an unverified observation into a mechanism, law, novelty claim, or physical contribution.
  • Do not solve a different, easier problem merely because it makes a cleaner model demo.
  • Prefer a small mechanism that can be disproved over a broad system with many loosely connected innovations.
  • Do not promise expensive training, external review, upload, or publication without the required authorization.

Report the frozen problem anchor, method chain, rejected additions, claim-to-validation table, residual risks, and the next Stage 1E action.

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.