Skills Guide

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.

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 I, plus Z_cross_cutting.
• J-level proofread and polishing entrypoints live at the workflow layer today; they are not a separate top-level skill stage in the internal registry.
• 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	Typical user intent
A_framing	topic framing, questions, theory, gap, venue	"What exactly is my contribution?"
B_literature	search, screen, extract, cite, map	"What does the literature say, and how do I build a corpus?"
C_design	design, variables, robustness, datasets	"How should this study be designed and operationalized?"
D_ethics	IRB, privacy, governance	"What ethics and data-protection materials do I need?"
E_synthesis	evidence synthesis, quality, bias	"How do I combine and rate evidence?"
F_writing	manuscript building, tables, figures, results writing	"How do I turn analysis into publishable text?"
G_compliance	reporting checklists, tone, PRISMA	"Is this compliant and submission-ready?"
H_submission	submission package, rebuttal, review simulation	"How do I package, defend, and stress-test the paper?"
I_code	academic code, stats, reproducibility	"How do I implement and verify research code?"
Z_cross_cutting	metadata, model collaboration, self-critique	"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	Use it when	Typical outcome
question-refiner	the topic is still vague or too broad	refined RQs, scope, inclusion/exclusion, search terms
hypothesis-generator	you need explicit hypotheses or propositions	testable hypotheses with mechanisms and boundaries
theory-mapper	you need a conceptual model or theory scaffold	theory map or Mermaid conceptual diagram
gap-analyzer	you need to justify novelty from prior work	prioritized gap analysis
venue-analyzer	you already know the paper direction and need venue fit	venue-fit memo, constraints, reviewer expectations

B. Literature

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

Skill	Use it when	Typical outcome
academic-searcher	you need reproducible search design and retrieval	search query plan, search results, search log
paper-screener	you need inclusion/exclusion decisions	screening log and PRISMA-ready counts
paper-extractor	you need structured notes from included papers	extraction table and per-paper notes
citation-snowballer	seed papers are known but coverage is incomplete	forward/backward citation expansion log
fulltext-fetcher	you have candidate papers but not the full text	full-text status and retrieval record
citation-formatter	references need to be normalized for writing	bibliography, citekeys, BibTeX-style outputs
concept-extractor	search concepts and controlled vocabulary need refinement	concept map and Boolean term set
literature-mapper	you need a taxonomy of streams, mechanisms, or clusters	literature map and open-problem structure
reference-manager-bridge	you need to exchange references with Zotero or similar tools	RIS/CSLJSON export or synced bibliography

C. Design

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

Skill	Use it when	Typical outcome
study-designer	you need the main study design package	design spec, analysis plan, instruments, prereg handoff
rival-hypothesis-designer	you need to stress-test the design against alternatives	rival-explanation matrix
robustness-planner	the empirical design needs sensitivity or robustness planning	robustness and sensitivity plan
dataset-finder	data availability is uncertain	dataset feasibility and access plan
variable-constructor	constructs must be turned into auditable variables	variable specification and coding rules

D. Ethics

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

Skill	Use it when	Typical outcome
ethics-irb-helper	you need the formal ethics package	IRB-facing materials, consent, recruitment, governance
deidentification-planner	privacy protection needs a technical plan	deidentification/privacy control plan

E. Synthesis

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

Skill	Use it when	Typical outcome
evidence-synthesizer	you need narrative synthesis or meta-analysis	synthesis narrative or pooled evidence result
quality-assessor	you need risk-of-bias and certainty judgments	quality/risk-of-bias assessment
publication-bias-checker	you need to test for publication bias	publication-bias report

F. Writing

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

Skill	Use it when	Typical outcome
manuscript-architect	you need the overall paper structure or section drafting plan	outline, section plan, draft spine
analysis-interpreter	you need results text that preserves uncertainty	bounded results narrative
effect-size-interpreter	coefficients need substantive interpretation	effect-size translation in human terms
table-generator	results must become paper-ready tables	publication-ready tables
figure-specifier	you need exact figure logic before plotting	figure specification and code guidance
meta-optimizer	title, abstract, and keywords need discoverability polish	optimized title/abstract/keywords

G. Compliance

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

Skill	Use it when	Typical outcome
prisma-checker	the review follows PRISMA-style evidence reporting	PRISMA completeness report
reporting-checker	you need CONSORT/STROBE/COREQ/SRQR/TRIPOD-style checking	reporting-guideline coverage report
tone-normalizer	text is too soft, too fluffy, or too absolute	concise academic tone revision log

H. Submission

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

Skill	Use it when	Typical outcome
submission-packager	journal/package artifacts need assembling	cover letter, disclosures, supplements bundle
rebuttal-assistant	reviewer comments need structured responses	point-by-point rebuttal matrix
peer-review-simulation	you want independent reviewer-style stress tests	multi-persona review memo
fatal-flaw-detector	you want a harsh desk-reject risk scan	fatal-flaw analysis
reviewer-empathy-checker	responses are technically correct but sound defensive	reviewer-response tone check

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	Use it when	Typical outcome
code-builder	you need to turn a method into executable research code	implementation package aligned to academic method
data-cleaning-planner	raw data cleaning must become an auditable procedure	cleaning rules and transformation plan
data-merge-planner	multiple data sources must be joined safely	merge strategy and provenance controls
code-specification	coding must start from a low-freedom contract	code/code_specification.md
code-planning	execution needs zero-decision, stepwise planning	code/plan.md
code-execution	planned implementation and profiling must be logged	code/performance_profile.md
code-review	a second pass must audit logic and statistical validity	code/code_review.md
reproducibility-auditor	reruns, seeds, and environment evidence must be checked	code/reproducibility_audit.md
stats-engine	you need modeling and diagnostics rather than general coding	analysis/stats_report.md

Z. Cross-Cutting

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

Skill	Use it when	Typical outcome
metadata-enricher	metadata across notes, references, and outputs is inconsistent	normalized DOI/author/venue/year metadata
model-collaborator	you need Codex, Claude, and Gemini to cross-check or shard work	multi-model execution and review plan
self-critique	you want structured red-teaming and revision pressure	critique questions and revision loop guidance

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
• 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