Skip to content

SKILL.md

Latest commit

 

History

History
221 lines (177 loc) · 11.1 KB

SKILL.md

File metadata and controls

221 lines (177 loc) · 11.1 KB
Raw
name kaic-research-skills
description Use when supporting research-paper statistics, data preparation, statistical interpretation, or manuscript reporting, especially for basic analysis, R, SPSS, PROCESS, mediation, moderation, propensity score matching, complex survey analysis, Korea Community Health Survey (KCHS, 지역사회건강조사) analysis in combination with kaic-chs-analysis, verified references, occupational-therapy research, or health-science research.

KAIC Research Skills

Overview

Support research-paper statistics, data preparation, interpretation, and manuscript reporting, with occupational-therapy and health-science research as the core domain specialization. Default to Korean for research consultation, statistical interpretation, analysis notes, and manuscript drafts unless the user explicitly requests another language.

Working Defaults

  • Prefer the data guide, codebook, instrument manual, and user-provided study rules over generic assumptions.
  • Prefer R for direct local analysis. Provide executable SPSS Syntax and PROCESS macro Syntax when the user wants SPSS workflows.
  • Ask only when a missing fact can materially change data preparation, statistical validity, or interpretation. For basic analyses with adequate inputs, proceed and state limited assumptions.
  • Separate Statistical interpretation from Manuscript reporting whenever results are interpreted or written for a paper.
  • Treat raw variable names as evidence. Preserve raw variables, create analysis variables explicitly, and record recoding, scoring, exclusions, and derived variables.
  • When raw names are Korean, contain spaces, or are otherwise awkward for code, preserve them and prefer a raw-to-analysis variable map with concise English aliases for sustained analyses. Use safe quoting/backticks when raw names must appear in R code.

Codex Model Guidance

When GPT-5.6 tier selection is available in Codex:

  • Recommend gpt-5.6-sol with high reasoning for study planning, analysis-critical data preparation, statistical modeling, result interpretation, and manuscript-critical work.
  • Reserve max for an integrated audit spanning study design, variable mapping, analysis specification, model sequence, sensitivity analyses, reference support, and manuscript logic.
  • Use gpt-5.6-terra with high reasoning for routine formatting, table cleanup, or revision after the analysis and source checks are complete.
  • Treat this guidance as advisory. Codex does not switch models automatically, and model choice never replaces executing the analysis or checking source guides, codebooks, verified references, and actual output.

Routing Priority

  • Use this skill as the default entry point for research-paper statistics, analysis preparation, statistical interpretation, or manuscript reporting.
  • When a request mentions KCHS, CHS, 지역사회건강조사, KDCA community health survey raw data, or variables such as kstrata, SPOT_NO, or wt_p, combine this skill with kaic-chs-analysis. Let kaic-chs-analysis govern KCHS design variables, missing codes, scoring, SPSS complex-samples syntax, and PROCESS limitations; use this skill for study planning, interpretation, manuscript reporting, references, and APA tables.
  • Prioritize this skill for statistical research work even when the study topic is outside occupational therapy or health science, especially for basic analysis, R, SPSS, PROCESS, propensity score matching, or complex surveys.
  • Do not treat this skill as the default for generic web research, market or technology research, or general writing that is not tied to a research paper.
  • Use or combine a more specific skill when the primary request is research ethics or integrity, generic preliminary research, general academic artifact review, or a dedicated Korean humanization pass.
  • Honor an explicit user request to use kaic-research-skills when it does not conflict with higher-priority safety or integrity requirements.

Integrity Rules

  • Never invent a paper, DOI, author list, title, journal, statistic, result, scoring rule, value set, dataset rule, or analysis output.
  • Verify references before presenting them as usable manuscript references. Include DOI information when it is verified and relevant. Mark missing or unverified DOI metadata plainly.
  • Do not infer missing-value codes, scale directions, survey weights, strata, clusters, treatment definitions, or outcome event coding when the guide or user context is required.
  • Do not rewrite observational findings into causal claims. Distinguish association, statistical indirect effect, moderation, matching-adjusted comparison, and causal effect.
  • Keep actual research data, restricted guides, identifying files, credentials, and manuscript-sensitive project artifacts out of this skill repository.

Reference Verification

Applies to: Manuscript mode (Introduction, Discussion) and Reference mode. Do NOT apply to Methods, Results, statistical analysis, or code generation modes.

Pool-First Procedure

When the user provides research results and requests Introduction or Discussion writing, build a verified reference pool before writing any prose.

Phase 1 — Build pool

  1. Extract key topics and keywords from the provided results.
  2. Search PubMed MCP for each topic (primary source for health/medical/OT).
  3. If PubMed returns fewer than 3 verified references for a topic, call Crossref API: https://api.crossref.org/works?query=<keywords>&rows=5 No API key required.
  4. If combined PubMed and Crossref results are still fewer than 3, call Semantic Scholar: https://api.semanticscholar.org/graph/v1/paper/search?query=<keywords>&fields=title,authors,year,externalIds No API key required.
  5. Add a reference to the pool only when the DOI is confirmed and title, author list, and year are complete metadata. Confirmed means the DOI is returned as a primary identifier in the PubMed or Crossref response, or resolves via https://doi.org/<DOI> without error.
  6. Report pool size before writing: "PubMed N건, Crossref N건, Semantic Scholar N건 — 검증된 문헌 N개 확보"

Phase 2 — Write

  • Cite only pool references. APA in-text: (Author, Year) or Author (Year).
  • If a claim requires a topic with 0 pool references, note inline: [검증된 문헌 없음 — 이 주제 탐색 결과 0건]
  • Never fabricate, guess, or use an unverified reference under any circumstances.

Phase 3 — Reference list

  • List only references actually cited in the body.
  • Full APA format with a DOI line for every entry.
  • Sort alphabetically by first author surname.
  • Example entry: 
    Kim, J., & Lee, S. (2022). Title of the article. Journal Name, 45(3), 123–135.
https://doi.org/10.xxxx/xxxxx

User-Supplied References

If the user provides a reference manually, verify it through the same PubMed → Crossref → Semantic Scholar pipeline before using it. Do not include user-supplied references in the pool without confirmed DOI and complete metadata.

Codex Compatibility

PubMed MCP is available in Claude Code sessions. In Codex, use available web search tools to query PubMed (https://pubmed.ncbi.nlm.nih.gov/?term=) as a fallback. Crossref and Semantic Scholar are plain HTTPS calls available in both environments.

Mode Routing

Read only the references needed for the active mode.

Mode Use when Read
Explore The user asks what a dataset can support or how to plan a study. references/explore-and-analysis-planning.md
Data Preparation The user needs recoding, scoring, derived variables, cleaning, or an analysis dataset. references/data-preparation-and-derived-variables.md
Run The user wants direct analysis of provided data. references/statistical-analysis.md and the relevant focused reference
Code The user wants paste-ready SPSS Syntax, PROCESS Syntax, or R code. references/spss-syntax-and-process.md or references/r-analysis-workflows.md
Interpret The user supplies SPSS, PROCESS, R, or result-table output. references/result-interpretation-and-reporting.md
Manuscript The user wants Methods, Results, Introduction, Discussion, or manuscript revision. references/manuscript-writing.md
Reference The user wants citations, DOI checks, APA references, or journal-style references. references/references-and-journal-styles.md
Table The user pastes statistical output and requests an APA-formatted table as a .docx file. references/apa-table-formatting.md

For mode selection details and common mode combinations, read references/modes-and-routing.md.

Analysis Routing

Use references/statistical-analysis.md for basic tests, regression families, hierarchical regression, mediation, moderation, Table 1, and output structure. Also read:

  • references/propensity-score-matching.md for propensity score matching.
  • references/complex-survey-analysis.md for survey weights, strata, clusters, survey descriptive statistics, or survey regression.
  • kaic-chs-analysis plus its relevant references for KCHS/지역사회건강조사 raw-data work; do not substitute generic survey assumptions for KCHS-specific rules.
  • references/data-preparation-and-derived-variables.md before any analysis that requires scoring, reverse coding, categorization, eligibility filtering, missing-code conversion, or derived variables.

Analysis Handoff

For substantial direct analysis or data preparation, produce a compact record of what was done. Use references/analysis-logs-and-handoff.md when a paper-facing analysis, multi-step preparation, cross-device continuation, or another analyst handoff is likely.

Manuscript Language

  • Write Korean academic prose that is precise, scoped, and readable. Avoid inflated filler, translationese, needless passive constructions, and repeated generic phrases.
  • Use English manuscript prose only when requested or when the supplied artifact is explicitly English.
  • Do not copy humanize-korean guidance into this skill. If the user asks for a dedicated Korean humanization pass or the draft needs that workflow, use the installed humanize-korean skill separately.

Bundled Templates

Use scripts as editable templates, not blind automation:

  • scripts/inspect_csv.R: profile a CSV before analysis.
  • scripts/basic_analysis_template.R: basic descriptive tests.
  • scripts/regression_template.R: linear, logistic, and hierarchical models.
  • scripts/table1_template.R: Table 1 skeleton.
  • scripts/psm_template.R: propensity score matching skeleton.
  • scripts/survey_template.R: complex survey skeleton.
  • scripts/apa_table_template.R: APA-formatted Word table builder (flextable + officer).

Output Contract

For analysis or interpretation tasks, prefer this compact structure when it fits:

  1. Inputs and assumptions
  2. Data preparation
  3. Statistical interpretation
  4. Manuscript reporting
  5. Reproducibility note or Analysis log when warranted

For code-only requests, put the paste-ready code first and keep explanatory notes focused on values the user must verify before running it.