76 lines (59 loc) · 5 KB

Mode Registry

Single source of truth for all modes across the ARS suite. 27 modes across 4 skills.

When adding or modifying modes, update this file first — SKILL.md files and CLAUDE.md should reference this registry.

Last updated: v3.19.0 (2026-06-21)

deep-research (8 modes)

Mode	Spectrum	Output	Oversight	Triggers
`full`	Balanced	APA 7.0 report, 3,000-8,000 words	High	"research [topic]", "deep research", "academic analysis"
`quick`	Fidelity	Research brief, 500-1,500 words	Medium	"quick brief", "30 minute summary", "quick research"
`review`	Balanced	Reviewer report on provided text	High	"review this paper", "evaluate this paper", "assess this source"
`lit-review`	Fidelity	Annotated bibliography + synthesis	Medium	"literature review", "annotated bibliography"
`three-way-scan`	Fidelity	WHY/HOW/WHAT paper shortlist + cross-paper synthesis	Low	"WHY HOW WHAT papers", "3W literature scan", "compare these papers"
`fact-check`	Fidelity	Claim-by-claim verification report	Medium	"verify claims", "fact-check", "evidence verification"
`socratic`	Originality	Research Plan Summary + INSIGHT collection	Very High	"guide my research", "help me think through", "I'm not sure what to research"
`systematic-review`	Fidelity	PRISMA 2020 report, 5,000-15,000 words	Medium	"systematic review", "meta-analysis", "PRISMA"

academic-paper (11 modes)

Mode	Spectrum	Output	Oversight	Triggers
`full`	Balanced	Complete paper draft (IMRaD or domain-appropriate)	High	"write a paper", "academic paper", "research paper"
`plan`	Originality	Chapter Plan + INSIGHT collection (Socratic)	Very High	"guide my paper", "help me plan", "step by step paper"
`outline-only`	Balanced	Detailed outline + evidence map	High	"paper outline", "just need an outline"
`revision`	Fidelity	Revised draft + point-by-point R&R responses	High	"revise paper", "incorporate reviewer feedback"
`revision-coach`	Balanced	Revision Roadmap + Response Letter Skeleton	Medium	"parse reviews", "I got reviewer comments"
`abstract-only`	Fidelity	English abstract (150-300 words, structured) + 5-7 keywords	Medium	"write abstract"
`lit-review`	Fidelity	Annotated bibliography in paper format	Medium	"literature review paper", "write a lit review"
`format-convert`	Fidelity	Formatted document (LaTeX/DOCX-via-Pandoc/PDF/MD)	Low	"convert to LaTeX", "convert citations to [format]"
`citation-check`	Fidelity	Citation error report	Low	"check citations", "verify references"
`disclosure`	Fidelity	Venue-specific AI-usage disclosure statement	Low	"AI disclosure for [venue]", "generate AI usage statement"
`rebuttal-audit`	Fidelity	Advisory QA of an existing rebuttal draft (per-comment coverage + gaps + risk flags); no generation; no Schema 11 emission	Low	"audit my response", "check my rebuttal", "did I miss any reviewer comment"

academic-paper-reviewer (6 modes)

Mode	Spectrum	Output	Oversight	Triggers
`full`	Balanced	5 review reports + Editorial Decision + Revision Roadmap	High	"review paper", "peer review", "manuscript review"
`re-review`	Fidelity	Revision verification checklist + residual issues	Medium	"check revisions", "verification review"
`quick`	Fidelity	EIC quick assessment + key issues list	Low	"quick review", "quick look"
`methodology-focus`	Fidelity	In-depth methodology review	Medium	"check methodology", "focus on methods"
`guided`	Originality	Socratic issue-by-issue dialogue	Very High	"guide me to improve", "walk me through issues"
`calibration`	Fidelity	Calibration Report (FNR/FPR/AUC) + confidence disclosure	Medium	"calibrate reviewer", "measure reviewer accuracy"

academic-pipeline (1 orchestrator + 1 resume mode)

Mode	Spectrum	Output	Oversight	Triggers
(pipeline)	Balanced	10-stage orchestrated workflow	Very High	"academic pipeline", "research to paper", "full paper workflow"
`resume_from_passport=<hash>`	Fidelity	Resume a prior pipeline run from a Material Passport reset boundary. Opt-in (`ARS_PASSPORT_RESET=1`). See `academic-pipeline/references/passport_as_reset_boundary.md`.	High	"resume from passport", "continue pipeline from reset boundary"

Summary

Metric	Count
Total modes	27
Fidelity	16 (59%)
Balanced	7 (26%)
Originality	4 (15%)

Oversight levels

Level	Meaning
Very High	User-led dialogue or mandatory checkpoints at every stage
High	User confirms key decisions (RQ, outline, configuration)
Medium	Structured format with limited decision points
Low	Mechanical/template-driven, minimal human input