Agent Skills: IMRaD Skill

Use when producing research-style outputs that should follow IMRaD (Introduction, Methods, Results, Discussion).

UncategorizedID: narumiruna/agent-skills/imrad-writer

Install this agent skill to your local

pnpm dlx add-skill https://github.com/narumiruna/agent-skills/tree/HEAD/skills/imrad-writer

Skill Files

Browse the full folder contents for imrad-writer.

Download Skill

Loading file tree…

skills/imrad-writer/SKILL.md

Skill Metadata

Name
imrad-writer
Description
Use when producing research-style outputs that should follow IMRaD (Introduction, Methods, Results, Discussion).

IMRaD Skill

Purpose

Generate outputs using the IMRaD structure by default:

  • Introduction
  • Methods
  • Results
  • Discussion

IMRaD is a standardized structure widely used in scientific writing to present research logically and clearly.

When to use

Use this skill when the user:

  • Asks for research-style writing
  • Needs structured analysis or reports
  • Wants systematic reasoning (problem -> method -> result -> interpretation)
  • Requests academic, technical, or experimental explanations

When NOT to use

Do NOT use this skill when:

  • The task is simple explanation or Q&A
  • The user asks for tutorials or step-by-step guides
  • The task is brainstorming or ideation
  • No clear "method -> result" structure exists and one cannot be reasonably inferred

Heuristic (quick decision rule)

Use IMRaD only if ALL are present:

  • A clearly defined question or problem
  • A describable method or approach exists or can be reasonably inferred
  • Observable or derivable results exist or can be explicitly scoped as inferred, hypothetical, expected, or simulated
  • A need to interpret those results

Handling Incomplete Inputs

If the input lacks explicit methods or results, the agent MUST:

  • Infer a reasonable method only when the task is analytical or design-oriented
  • State assumptions explicitly in Methods
  • Generate logically consistent Results based on those assumptions
  • Label non-empirical results explicitly as inferred, hypothetical, expected, or simulated

Truthfulness Constraints

The agent MUST NOT fabricate empirical data, experiments, citations, or observations.

If actual results are unavailable, the Results section MUST present only:

  • Inferred results
  • Hypothetical results
  • Expected results
  • Simulated results

Such results MUST be labeled explicitly.

Output Requirements

The output MUST:

  • Follow IMRaD section order exactly
  • Use clear section headers
  • Be concise, precise, and non-redundant
  • Avoid mixing content across sections
  • Maintain logical flow

Section Specifications

1. Introduction

  • Define background and context
  • Identify the problem or research question
  • State objective or hypothesis

Constraint:

  • No methods, results, or conclusions here

2. Methods

  • Describe approach, methodology, or process
  • Include assumptions, tools, or data sources
  • Ensure reproducibility when applicable
  • Include assumptions explicitly when the input is incomplete

Constraint:

  • No interpretation or results

3. Results

  • Present findings, outputs, or observations
  • Use structured bullet points if helpful
  • Stay objective and factual

Constraint:

  • No interpretation, explanation, or implication
  • Only describe observations or outputs

4. Discussion

  • Interpret results
  • Explain implications and significance
  • Compare with expectations or prior knowledge
  • Explicitly answer the research question from Introduction
  • Mention limitations and possible next steps

Style Constraints

  • Language MUST be concise and precise
  • Avoid narrative storytelling
  • Prefer structured bullets over long paragraphs
  • No filler or meta commentary
  • No repetition across sections

Optional Extensions

Include ONLY if explicitly requested:

  • Abstract (summary of all sections)
  • Conclusion (can be merged into Discussion)

Example Invocation

User input: "Analyze why async does not solve race conditions in file writes"

Expected structure:

Introduction

Explain the concurrency problem in file writes and define the research question.

Methods

Examine async execution, shared-state access, and file write behavior under concurrent calls.

Results

  • Async allows concurrent scheduling
  • It does not prevent multiple writers from targeting the same key
  • Without locking, last-write-wins behavior remains possible

Discussion

Interpret why async changes execution style but does not solve cross-process or shared-resource race conditions.

Failure Modes (MUST avoid)

  • Mixing Results with Discussion
  • Introducing conclusions before Results
  • Writing unstructured essays
  • Skipping sections