Skills Guide

Auto-generated from skills/registry.yaml by python3 scripts/generate_skill_docs.py. Do not edit this file by hand.

This page is the user-facing map of what the skills/ layer contains.

It is meant to answer questions such as:

• Which part of the system handles my current research problem?
• What does each stage actually contain?
• Which skills are canonical and auto-routed?
• Which markdown cards are supplemental helpers or mirrors rather than primary routed skills?

Canonical Source

The canonical routed skill list lives in skills/registry.yaml. The tables below summarize that registry for human readers. User-facing surfaces may read display_name, when_to_use, summary_zh, display_name_zh, and when_to_use_zh directly from that registry.

How Users Should Read The Skills Layer

• A workflow command such as /paper or /code-build is the entry UX.
• A Task ID such as B2, F3, or I6 is the contract-level unit of work.
• A skill is the reusable execution behavior that the orchestrator injects behind the scenes through required_skills and required_skill_cards.

In other words, most users should not manually choose raw markdown skill files one by one. You usually choose:

1. a workflow entrypoint, or
2. a Task ID via task-plan / task-run.

Then the system decides which skills to load.

If you need exact runtime flags, use CLI Reference. If you need to understand how agents and skills interact at runtime, use Agent + Skill Collaboration. If you need to modify the system, use Extend Research Skills. If you want scenario-driven routes such as "systematic review", "methods paper", or "rebuttal prep", use Task Recipes.

Important Boundaries

• The current internal skill registry covers stages A through K except there is no routed top-level L or beyond; J_proofread, K_presentation, and Z_cross_cutting are first-class registry stages.
• Some markdown files under skills/ are supplemental cards or mirror copies for the Stage-I code lane. They are documented below, but they are not all separate routed skills.

Stage Overview

Stage	Focus	Skill count	Typical user intent
A_framing	topic framing, questions, theory, gap, venue	6	"What exactly is my contribution?"
B_literature	search, screen, extract, cite, map	9	"What does the literature say, and how do I build a corpus?"
C_design	design, variables, robustness, datasets	9	"How should this study be designed and operationalized?"
D_ethics	IRB, privacy, governance	3	"What ethics and data-protection materials do I need?"
E_synthesis	evidence synthesis, quality, bias	5	"How do I combine and rate evidence?"
F_writing	manuscript building, tables, figures, results writing	7	"How do I turn analysis into publishable text?"
G_compliance	reporting checklists, tone, PRISMA	3	"Is this compliant and submission-ready?"
J_proofread	AI detection, humanization, similarity, final polish	4	"How do I de-AI and finalize the manuscript?"
H_submission	submission package, rebuttal, review simulation	7	"How do I package, defend, and stress-test the paper?"
I_code	academic code, stats, reproducibility	10	"How do I implement and verify research code?"
K_presentation	academic talks, slide planning, Slidev, Beamer	4	"How do I turn the paper into a defensible talk?"
Z_cross_cutting	metadata, model collaboration, self-critique	4	"How do I improve quality across stages?"

Canonical Skills By Stage

A. Framing

Use Stage A when you are still defining the research question, contribution, theory anchor, or venue positioning.

Skill	Display Name	When to use	Produces
question-refiner	Question Refiner	Use when you need to transform vague topics into structured rqs via pico/peo + finer evaluation.	RQSet
contribution-crafter	Contribution Crafter	When framing the pitch for a manuscript introduction	ContributionStatement
hypothesis-generator	Hypothesis Generator	Use when you need to translate rqs into testable hypotheses with mechanisms and boundary conditions.	HypothesisSet
theory-mapper	Theory Mapper	Use when you need to map theoretical concepts, relationships, and frameworks with mermaid diagrams.	TheoreticalFramework
gap-analyzer	Gap Analyzer	Use when you need to identify and categorize research gaps using 5-type taxonomy with finer prioritization.	GapAnalysis
venue-analyzer	Venue Analyzer	At the very start of a project (affects RQ scope, methods choice, page limits)	VenueAnalysis

B. Literature

Use Stage B when you are building or maintaining the literature base for a topic, especially systematic or reproducible reviews.

Skill	Display Name	When to use	Produces
academic-searcher	Academic Searcher	Use when you need to own query design, provider-backed retrieval, and deduplicated search outputs as one top-level search skill.	SearchQueryPlan, SearchResults, SearchLog
paper-screener	Paper Screener	Use when you need to two-stage prisma-compliant systematic screening with decision logging.	ScreeningDecisionLog, PRISMAFlowData
paper-extractor	Paper Extractor	Use when you need to extract structured theory, method, data, findings, and limitation slots into notes and rollups.	ExtractionTable, PaperNotes
citation-snowballer	Citation Snowballer	Use when you need to forward/backward citation tracing to expand corpus and find seminal works.	SnowballLog
fulltext-fetcher	Full-text Fetcher	Use when you need to retrieve full-text pdfs via oa channels with prisma-compliant status tracking.	FullTextStatus
citation-formatter	Citation Formatter	Use when you need to format citations in apa/mla/chicago/ieee/bibtex with consistent citekeys.	Bibliography
concept-extractor	Concept Extractor	After research questions (A1) are drafted but before search execution (B1)	ConceptMap
literature-mapper	Literature Mapper	After paper extraction (B2) when you have ≥15 papers with structured notes	LiteratureMap
reference-manager-bridge	Reference Manager Bridge	Use when you need to export/import references between research system and zotero/mendeley/endnote.	Bibliography, RISExport, CSLJSONExport

C. Design

Use Stage C when the question is already clear and the next problem is design validity, data feasibility, and operationalization.

Skill	Display Name	When to use	Produces
study-designer	Study Designer	After /find-gap and /build-framework, when you want to run an empirical, qualitative, or mixed-methods study.	DesignSpec, AnalysisPlan, DataManagementPlan, Instruments, Preregistration
rival-hypothesis-designer	Rival Hypothesis Designer	After hypotheses (A1_5) and study design (C1) are drafted	RivalHypotheses
robustness-planner	Robustness Planner	After analysis plan (C3) is drafted	RobustnessPlan
dataset-finder	Dataset Finder	After study design specifies what data is needed	DatasetPlan
variable-constructor	Variable Constructor	After study design is finalized	VariableSpec
data-dictionary-builder	Data Dictionary Builder	After study design is finalized and instruments are selected	DataDictionary
data-management-plan	Data Management Plan	Use when you need to generate fair-compliant data management plans specifying storage, backup, retention, sharing, and archival.	DataManagementPlan
prereg-writer	Pre-registration Writer	After study design and analysis plan are finalized	Preregistration
variable-operationalizer	Variable Operationalizer	After theoretical framework is established	OperationalizationMap

D. Ethics

Use Stage D when the study touches human participants, sensitive data, governance, or data-release constraints.

Skill	Display Name	When to use	Produces
ethics-irb-helper	Ethics & IRB Helper	Before data collection involving human participants or sensitive data	EthicsPackage
statement-generator	Statement Generator	When finalizing a manuscript for submission and ensuring compliance with journal statement requirements (e.g., PLOS, Nature, Elsevier).	Manuscript
deidentification-planner	Deidentification Planner	When handling any dataset containing identifiable information	DeidentificationPlan

E. Synthesis

Use Stage E when the evidence base already exists and the task is to combine, rate, or stress-test that evidence.

Skill	Display Name	When to use	Produces
effect-size-calculator	Effect Size Calculator	When preparing data for a quantitative meta-analysis	EffectSizeTable, AnalysisCode
evidence-synthesizer	Evidence Synthesizer	Use after you have:	EvidenceTable, SynthesisMatrix
quality-assessor	Quality Assessor	Use when you need to assess risk of bias and certainty via rob 2, robins-i, grade.	QualityTable, GRADESummary
publication-bias-checker	Publication Bias Checker	After completing meta-analysis (E3) with ≥5 studies	PublicationBiasReport
qualitative-coding	Qualitative Coding	Use after raw qualitative data has been collected or extracted, but before the final synthesis narrative is drafted. Ideal for Grounded Theory, Thematic Analysis, or Interpretative Phenomenological Analysis (IPA).	DataDictionary, ThematicCodebook

F. Writing

Use Stage F when the main question is turning evidence and analysis into sections, tables, figures, and readable claims.

Skill	Display Name	When to use	Produces
manuscript-architect	Manuscript Architect	You want to draft a paper from an existing RESEARCH/[topic]/ project folder (empirical study or systematic review).	ManuscriptOutline, Manuscript, ClaimGraph, FiguresTablesPlan
analysis-interpreter	Analysis Interpreter	After primary analysis is complete	ResultInterpretation
effect-size-interpreter	Effect Size Interpreter	After primary analysis is complete	EffectInterpretation
table-generator	Table Generator	After primary analysis is complete	FormattedTables
figure-specifier	Figure Specifier	After primary analysis is complete	FigureSpecs
meta-optimizer	Meta Optimizer	After manuscript draft is complete (F2/F3/F4)	MetaOptimization
discussion-writer	Discussion Writer	Use after the Results section has been drafted and the primary findings are finalized. Provide the core contribution statement and literature framing to ensure alignment.	DiscussionDraft, StorySpine

G. Compliance

Use Stage G when the paper exists and now needs formal checklist coverage, tone cleanup, or reporting verification.

Skill	Display Name	When to use	Produces
prisma-checker	PRISMA Checker	Use when you need to verify prisma 2020 flow diagram and all 27 checklist items.	PRISMAChecklist
reporting-checker	Reporting Checker	Before submission (final quality assurance pass)	ReportingChecklist
tone-normalizer	Tone Normalizer	After manuscript draft is complete (but before final submission)	ToneNormalization

J. Proofread

Use Stage J when the draft is substantively complete and needs AI-fingerprint review, human-voice rewriting, similarity screening, or final proofreading before submission.

Skill	Display Name	When to use	Produces
ai-fingerprint-scanner	AI Fingerprint Scanner	After the first complete manuscript draft (F2+)	AIDetectionReport
human-voice-rewriter	Human-Voice Rewriter	After J1 (AI fingerprint scan) identifies high/medium severity passages	HumanizedManuscript
similarity-checker	Similarity Checker	After J2 (human-voice rewrite) to ensure rewrites are original	SimilarityReport
final-proofreader	Final Proofreader	As the final step before submission (J4)	ProofreadChecklist

H. Submission

Use Stage H when the manuscript is near submission or already under review.

Skill	Display Name	When to use	Produces
submission-packager	Submission Packager	You have a near-final manuscript draft ready for submission	SubmissionPackage
rebuttal-assistant	Rebuttal Assistant	After receiving a Revise & Resubmit (R&R) decision	ResponseToReviewers, ResponseLetter
peer-review-simulation	Peer Review Simulation	Before submission (final red-team pass)	PeerReviewSimulation
fatal-flaw-detector	Fatal Flaw Detector	Before final submission (last quality gate)	FatalFlawAnalysis
reviewer-empathy-checker	Reviewer Empathy Checker	After drafting the response letter (from rebuttal-assistant H2) but BEFORE submitting	EmpathyCheck
credit-taxonomy-helper	CRediT Taxonomy Helper	During submission packaging (H1)	CRediTStatement
limitation-auditor	Limitation Auditor	Use during the late drafting or pre-submission phase when the Methods and Discussion sections are complete, to ensure the manuscript demonstrates scholarly humility and self-awareness before peer reviewers point out the flaws.	LimitationSection, MitigationStrategy

I. Code

Use Stage I for academic code, data workflows, statistical execution, and reproducibility. This lane is stricter than general engineering prompts.

The core strict sequence is:

1. code-specification
2. code-planning
3. code-execution
4. code-review
5. reproducibility-auditor

That sequence is what code-build --focus full is designed to reinforce.

Skill	Display Name	When to use	Produces
code-builder	Code Builder	Use when you need to convert academic methods into executable code with domain-profile library selection.	AnalysisCode
data-cleaning-planner	Data Cleaning Planner	After dataset is obtained, before any analysis	CleaningPlan
data-merge-planner	Data Merge Planner	When analysis requires combining multiple data sources	MergePlan
code-specification	Code Specification	Use when you need to generate strict opsx-style constraint sets before coding.	CodeSpec
code-planning	Code Planning	Use when you need to transform specs into parallelizable, zero-decision execution plans.	CodePlan
code-execution	Code Execution	Use when you need to execute code plans with cprofile performance profiling and optimization.	PerformanceProfile
code-review	Code Review	Use when you need to secondary model reviews code logic, security, and statistical validity.	CodeReview
reproducibility-auditor	Reproducibility Auditor	Use when you need to verify seeds, containerization, and fail-graceful contingencies.	ReproducibilityReport
release-packager	Release Packager	When preparing for manuscript submission (data/code availability statement)	ReleasePackage
stats-engine	Stats Engine	You need to execute modeling/testing for synthesis (E3/E3_5) or empirical results (F stage) with a clear report of assumptions and uncertainty.	StatsReport

K. Presentation

Use Stage K when the paper already exists and the next task is to turn it into a talk, seminar deck, or conference presentation.

Skill	Display Name	When to use	Produces
presentation-planner	Presentation Planner	After manuscript is substantially complete (or results are available)	PresentationPlan
slide-architect	Slide Architect	After presentation-planner has produced a slide blueprint	SlideDeckSpec
slidev-scholarly-builder	Slidev Scholarly Builder	When Slidev + scholarly is chosen as the output backend	SlidevDeck, BibTeXFile
beamer-builder	Beamer Builder	When LaTeX Beamer is the chosen backend	BeamerDeck, BibTeXFile

Z. Cross-Cutting

Use Stage Z when the need cuts across stages rather than belonging to one paper section.

Skill	Display Name	When to use	Produces
metadata-enricher	Metadata Enricher	Use when you need to normalize and enrich doi, venue, year, author metadata across artifacts.	Bibliography
academic-context-maintainer	Academic Context Maintainer	Use when you need a stage-aware academic state summary that preserves research question scope, locked methodological choices, stable findings, unresolved disputes, and decision rationale across long-running paper workflows.	ResearchStateSnapshot, ResearchDecisionLog
model-collaborator	Model Collaborator	Use when literature screening, peer review simulation, rebuttal drafting, qualitative coding, or code/statistics validation benefits from independent multi-model passes before synthesis.	CollaborationTrace
self-critique	Self-Critique	Use when you need to iterative red teaming against superficial reasoning, contradictions, and overclaiming.	CritiqueLog

Supplemental Cards And Mirror Files

Not every markdown file under skills/ is a primary routed skill.

Supplemental Manual Cards

These are useful reference or helper cards, but they are not all first-class entries in the current registry:

File	Role
skills/C_design/data-dictionary-builder.md	builds a structured data dictionary
skills/C_design/data-management-plan.md	writes FAIR-style data management plans
skills/C_design/prereg-writer.md	drafts preregistration materials
skills/C_design/variable-operationalizer.md	maps constructs to measurable variables
skills/H_submission/credit-taxonomy-helper.md	prepares CRediT contribution statements
skills/I_code/release-packager.md	packages code/data/environment for archival release

Stage-I Mirror Directories

These mirror the canonical Stage-I cards so prompts can stay close to the execution lane:

• skills/I_code/build/
• skills/I_code/planning/
• skills/I_code/run/
• skills/I_code/qa/

Treat the canonical top-level files under skills/I_code/ as the main reference unless you are editing the implementation details of the Stage-I lane itself.

Cross-Cutting Alias

skills/Z_cross_cutting/tone-normalizer.md is a cross-cutting alias to the canonical compliance-oriented tone normalization behavior at skills/G_compliance/tone-normalizer.md.

Domain Profiles

The base skill system stays generic. Domain specialization is injected at runtime through skills/domain-profiles/*.yaml.

Current shipped profiles include:

• biomedical
• business-management
• cs-ai
• ecology-environmental
• economics
• education
• epidemiology
• finance
• political-science
• psychology

Use domain profiles when:

• the default framing or design logic is too generic
• the code lane needs domain-specific diagnostics
• reporting or venue expectations differ by field

For example, the Stage-I code lane can load field-specific method checks through --domain.

Which Page Should You Use Next?

• Need command syntax: CLI Reference
• Need to understand layer boundaries: Conventions
• Need to understand runtime cooperation: Agent + Skill Collaboration
• Need to change or add skills: Extend Research Skills