sv-skills

Plugin sv-skills from marketplace supervisor-skills-plus in easygl1der/supervisor-skills-plus. Commands read from the repo's manifest — 20 days ago.

Skills (64)

• research-core

  Shared foundation loaded by every sv- skill: core rules that hold in any field, field and venue profiles, sourced checklists, the AI-collaboration rules, and project templates such as research.yaml and STATUS.md. Not a slash command. Use when another skill says to load research-core, or when a rule,

• sv-experiment-freeze

  Freezes what an experiment depends on before any model output is seen: the corpus, the sample and gold set, model and prompt versions with a pre-set acceptance bar, and the rule that disagreements become written rules rather than silent fixes. Use when setting up an experiment, building a gold set o

• sv-experiment-numbers

  Moves numbers from runs into the paper so each one is traceable: a numbers file generated by scripts, macros or tables built from it, a check that every number in the manuscript matches the file, and verified numbers recorded in STATUS before they reach the text. Use when numbers enter or change in

• sv-experiment-prereg

  Locks a confirmatory analysis before outcomes are seen: hypotheses, the estimand, the exact primary analysis, the decision rule, stopping and multiplicity are written down and frozen as a git commit, and an audit later proves the registration preceded every committed output and was never edited. Any

• sv-experiment-replication

  Prepares where data lives and the replication artifact: data placement and compute backend, a manifest of data files that can be rebuilt and verified, an anonymized artifact checked for identity leaks, and the mirror for double-blind review. Use when deciding data storage, packaging a replication ar

• sv-experiment-runs

  Keeps the runs log and the state files of a project's experiments: a row per run when it starts with command, code commit, data version, backend, status and outputs, failed runs recorded and never deleted, outputs never overwritten, and diffable state files as the authority over web dashboards. Use

• sv-experiment

  Entry point for experiments whose printed numbers must be traceable: freezing corpus, samples, gold sets and model or prompt versions, the runs log, numbers checked into the paper, and the replication artifact. Routes to one sv-experiment sub-command and says which. Use when setting up, running, rep

• sv-feedback-guidance

  Records expert guidance on a paper or experiment as a dated guidance record: the scenario shown, verbatim quotes, the interpretation marked as such, points to notice, actions with owners, due dates and done criteria, and the decision; later lists open, overdue and unevidenced actions. Use when advic

• sv-feedback-rebuttal

  Provides actionable peer-review rebuttal guidance, not finished rebuttal prose: inventories and classifies concerns, matches the reviewer mindset and strategy on ML venues, gives per-concern plans with evidence pointers, and picks a conference, journal or internal response format. Use when the user

• sv-feedback-revise

  Turns comments on a manuscript into tracked revisions: collects every comment from every channel, gives it an ID, triages it, applies one change per comment with the ID in the commit, keeps note markers until their owner closes them, verifies each claimed change against the actual diff, and fills th

• sv-feedback

  Entry point for handling feedback: turns comments from advisors, co-authors, reviewers or agents into tracked revisions with IDs, one change per comment and diff verification; records expert guidance with actions and follow-up; and plans peer-review rebuttals by reviewer mindset and strategy. Routes

• sv-figure-design

  Advises on the design of a technical paper's three core figures: the motivated example (Figure 1), the solution overview, and the experimental results. Identifies the figure type, recommends a paradigm, sketches the layout, guides labels and annotations, suggests a tool, and audits the universal and

• sv-figure-drawio

  Reconstructs reference images into high-fidelity, editable Draw.io files with rendered previews: native Draw.io elements carry text and structure, SVG covers simple icons that match the reference, and cropped or transparent PNGs preserve complex visuals. Use when the user wants a diagram image, rese

• sv-figure-explain

  Draws explanatory figures for learning: scene cards, four-panel comics, textbook and auto-laid-out DAGs, mind maps, explainer pages and steppers for proofs, algorithms and numeric processes, built from the vendored offline kit as HTML with SVG and KaTeX, rendered to PNG and inspected. Never invents

• sv-figure-html

  Produces SVG or HTML figures with the vendored offline kit: copy a template, edit its data block, render to PNG in headless Chrome, inspect, and save the HTML and PNG together. Covers explainer pages, steppers, DAGs and mind maps for papers, slides or the web, and figures in a Claude Design style. U

• sv-figure-plot

  Produces data and mathematical figures from code, saving the figure and the script together: Matplotlib or Seaborn in a reusable plotting script, Plotly for interactive views, vector output sized to the column, and numbers that trace to the project's numbers file. Applies the results figure guidance

• sv-figure-poster

  Plans and checks a research poster with the library's figure rules: one message per panel, figures reused from the paper under the universal design rules, a tool chosen from the tool matrix, and the archive figure rules audit. States plainly that no poster-specific rules exist yet. Use when the user

• sv-figure-tikz

  Builds TikZ and PGFPlots figures for LaTeX papers and decks in a render-and-inspect loop: a standalone render to PDF and PNG, inspection of the PNG for overlaps, clipping and legibility, a short review note, and done only when the inspection is recorded. Follows the universal figure rules. Use when

• sv-figure

  Entry point for every figure task: design advice for a paper's core figures, draw.io reconstruction of a reference image, plots from code, TikZ, HTML or SVG figures, posters, and explanatory figures for learning. Routes to one sv-figure sub-command and says which. Use when the user asks to design, d

• sv-idea-evaluate

  Evaluates a preliminary research idea in full: first impression and paper type, the fatal-flaws gate, empirical-study and theory routes, lifecycle and student-capability matching, five-dimension scoring (higher, faster, stronger, cheaper, broader), a paradigm-shift probe, feasibility, the integrity

• sv-idea-fatal-flaws

  Runs only the early gate of the idea evaluation: the audit against the ten canonical fatal flaws (no novelty against the closest work, wrong venue, wrong baseline, weak motivation, capability mismatch, unverifiable claim, ethical or data blocker, scope, solution hunting, no failure case), at most tw

• sv-idea

  Entry point for judging a research idea: a full evaluation with first impression, fatal-flaws gate, lifecycle and capability matching, five-dimension scoring, paradigm-shift probe, feasibility and a verdict, or only the fatal-flaws gate when time is short. Routes to one sv-idea sub-command and says

• sv-litreview-map

  Builds a literature map from what the user has read: quickread cards and study notes found through the reading root's index, an evolution map of read and cited-only papers with typed edges, a comparison table with sources, an evidence-backed gap list, and next reads by backward and forward citation

• sv-litreview-survey

  Runs a survey-grade literature investigation: checks search capability, freezes the research brief, scouts from several adversarial perspectives, verifies every citation, synthesizes evidence into a MECE taxonomy with in-sentence citations and calibrated hedges, and passes six quality gates before d

• sv-litreview

  Entry point for literature work: a survey-grade investigation with frozen questions, adversarial search, verified citations and a calibrated synthesis, or a literature map built from the cards and notes already read. Routes to one sv-litreview sub-command and says which. Use when the user needs a su

• sv-proposal-grant

  Plans and drafts a grant application for NSFC, NSF, ERC or NIH: the call sheet with verified limits, the proposal spine of questions, aims and feasibility, an evidence ledger, section map, timeline and risks, a draft in the funder's register, and a reviewer pass against the funder's criteria. Use wh

• sv-proposal-kaiti

  Plans and drafts a thesis proposal (kaiti report): the call sheet from the department's rules, the proposal spine of questions, aims and feasibility, an evidence ledger, section map, timeline and risks, a draft in the Chinese kaiti register with visible gap markers, and a reviewer pass. Use when the

• sv-proposal-statement

  Plans and drafts a research statement or a PhD research proposal: the spine of questions, aims and feasibility, an evidence ledger, section map and timeline, a draft in proposal register, and a reviewer pass. Use when the user needs a research statement for the job market or a fellowship, or a resea

• sv-proposal

  Entry point for research proposals: thesis proposals (kaiti reports), grant applications such as NSFC, NSF, ERC and NIH, and research statements or PhD plans, built from a spine, evidence ledger, timeline and risks before drafting. Routes to one sv-proposal sub-command and says which. Use when the u

• sv-quickread-card

  Writes the nine-section quick-read card for one paper: problem, core idea, a field-specific method block, results copied with locations, contributions tagged proved, empirical or conjectured, a claim-evidence table, limitations, and the link to the reader's research. Text through MinerU only; checke

• sv-quickread-style

  Writes the quick-read card and also dissects the paper's writing into one Paper Style Card (structure, rhetorical moves, sentence templates) in the sv-write-style-mine format, saved beside the card as a candidate for a venue profile. One card never becomes a rule. Use when the user wants a paper rea

• sv-quickread

  Entry point for quick reads of one paper: a one-page, five-minute card in nine fixed sections with sourced results, graded contributions and a claim-evidence table, optionally with a Style Card of its writing. Routes to one sv-quickread sub-command and says which. Use when the user hands over one pa

• sv-review-loop

  Runs a bounded multi-round check over a manuscript: research-writing fit first (claims, scope, hedges, numbers, threats), then venue and field style, then language through a de-AI audit, with a regression re-run of the research check. Applies only authorized edits, guards spans against reversion and

• sv-review-once

  Runs a read-only pre-submission review of a technical paper across five dimensions (macro logic, writing details, English grammar, LaTeX formatting, figure quality) with CRITICAL, MAJOR and MINOR findings, then the research-core threats, double-blind, artifact, venue-submission and integrity checkli

• sv-review

  Entry point for reviewing a manuscript before it leaves the group: a read- only pre-submission audit with sub-checklists and a submission-day runbook, or a bounded multi-round check of research fit, venue style and language with a round log. Routes to one sv-review sub-command and says which. Use wh

• sv-slides-talk

  Builds a 12 to 18 slide critical paper talk in Beamer: the paper's section tree mapped to slide groups, plus the critical slides (paper card with estimand and identification, ranked assumptions, results with sources and no causal verbs, claims against evidence, link to the project, reproducibility,

• sv-slides-weekly

  Builds the weekly-report Beamer deck: conclusions first, last week's next steps marked done, partial or dropped, results with sources, blockers, decisions needed, and next steps with owner, date and kill criterion; the progress variant groups the same spine by topic. Compiles, lints, renders and exp

• sv-slides

  Entry point for LaTeX Beamer decks for math and statistics group meetings: a weekly report that carries last week's next steps forward, or a critical paper talk with estimand, identification and reproducibility slides, compiled, linted and rendered. Routes to one sv-slides sub-command and says which

• sv-study-bridge

  Bridges something just read or learned to the current research project: maps the source setting onto the project, checks which assumptions transfer, gives one of five verdicts on whether and how to use it, and designs the cheapest validation experiment with expected results and a kill criterion fixe

• sv-study-chapter

  Works through a textbook chapter or lecture notes understanding-first: maps the chapter's concepts and their order, checks prerequisites against the learner profile with one short question each, runs the protocol per concept with restatement and diagnosis, and keeps book progress in the study vault.

• sv-study-drill

  Switches on one of the study drill options that are off by default: Socratic dialogue, Feynman explain-back, light self-test with confidence ratings, proof reconstruction with gaps for the learner, notes only, or Anki export of approved concept notes. High-confidence wrong answers go to the misconce

• sv-study-paper

  Works through one paper understanding-first: official PDF first through MinerU, the concepts the paper needs planned in dependency order, the protocol run per concept with restatement and diagnosis, formulas and numbers copied with locations, and the session saved to the paper folder and the study v

• sv-study-quick-explain

  Explains one concept understanding-first with no source material: plain principle, motivation, a minimal example, links to what the learner already knows, one or two pictures, then the learner's restatement, diagnosis against the expert mental model and correction. Uses the study vault's profile whe

• sv-study-visual-explain

  Draws the motivation or structure figure for the concept being studied, as the study protocol's picture step: picks a kit template, fills it from the session's content without inventing numbers, renders and inspects it, and saves the HTML and PNG next to the notes. A thin entry to the figure group's

• sv-study

  Entry point for understanding-first study: explains a concept, works through a paper or textbook chapter one idea at a time with the learner's restatement and diagnosis, draws explanatory figures, runs optional drills, and bridges learning to the project. Routes to one sv-study sub- command and says

• sv-supervisor-brief

  Gives a read-only session brief for a research project: the stage in research.yaml and the stage the repository signals suggest, the next deadline with its source, blockers, overdue guidance, open reviewer markers and records warnings, then routes to one to three skills with a reason each. Routes re

• sv-supervisor-harvest

  Turns a project's feedback tables, guidance records, decision records, review-loop logs, STATUS and git history, plus expert notes, into sourced and deduplicated rule candidates for research-core, classifies where each belongs, and proposes the change on a learn branch for the owner to approve. Also

• sv-supervisor-init

  Starts a research project: interviews the user one short block at a time (field, paper type, stage, venue and deadline with its official source, repositories and the one writable source of truth, data and compute, records and co-author note macros), writes research.yaml after showing the diff, and s

• sv-supervisor-records

  Keeps a research project's records outside the paper: STATUS with the current state only, TODO, numbered decision records, dated logs and group-meeting logs, sources of truth for paper, data and numbers with a sync guard, the notes lifecycle, de-accretion of docs, and an optional Linear mirror. Use

• sv-supervisor

  Entry point for running a research project: a start-up interview that writes research.yaml and docs, a session brief with stage, deadline, blockers and next skills, project records (STATUS, TODO, decisions, logs), and lessons harvested into research-core. Routes to one sv-supervisor sub- command and

• sv-write-critique

  Critiques a draft without editing it: runs the source-recall and verification checklist, the reader-logic checks, the logic-chain consistency checks when a skeleton exists, and the research-core writing checklist, then says whether each problem is language, structure, or a claim the evidence may not

• sv-write-humanize-audit

  Runs the report-only de-AI audit: S1 to S7 scores, the five tests (topic-sentence, movability, asymmetry, falsifier, deletion), a gate verdict, and a finding list where every finding carries an edit class for sv-review-loop. Never edits. Use when the user wants a de-AI gate before review or submissi

• sv-write-humanize-letter

  Removes machine-written texture from letters and applications: response letters, rebuttals and emails to advisors use the academic-response profile (version narration expected, one polite opener allowed); proposals and grant applications use the academic profile. Same protected-content contract as t

• sv-write-humanize-paper

  Removes machine-written texture from a paper, thesis chapter or section under the academic profile: diagnoses restatement, defensiveness, false symmetry and reporting hygiene as S1 to S7 against a claim ledger, then edits only under the contract that protects numbers, citations, terms and LaTeX. Use

• sv-write-humanize-report

  Removes machine-written texture from reports: weekly and progress reports, STATUS and other files under docs/ use the docs profile (clarity over voice); technical reports for outside readers use the academic profile. Same diagnosis and editing contract as the paper sub-command, so numbers, citations

• sv-write-humanize-slides

  Removes machine-written texture from slide decks: frame titles and bodies use the slides profile (one hedge per claim, a selling point stated once per deck, no slogan closers, title verbs at or below the evidence level), while Beamer note text uses the academic profile. Use when a Beamer or other de

• sv-write-intro

  Drafts the Introduction as flowing prose. Technique and new-problem papers follow the six-paragraph flowchart (background and running example, limitations, goal, challenges, solution overview, contributions); RQ-driven empirical studies follow the outsider-question route to research-question blocks.

• sv-write-polish

  Polishes existing academic prose while preserving the author's meaning: grammar and flow repair, idiomatic wording, section conventions, tone calibration against evidence strength, and fair description of prior work. Light touch by default, one pass only, and every edit that could change scientific

• sv-write-section

  Drafts publishable paper prose from the author's own materials, from one paragraph to a full manuscript, across STEM and non-STEM fields. Builds the claim ledger before prose, traces every factual claim to user input, verified retrieval or field common knowledge, and runs the citation verification l

• sv-write-skeleton

  Plans a paper's logic skeleton before any prose. Technique and new-problem papers get the thinking-template table and four self-consistency checks; benchmark papers get the five-pillar audit, the six-part Introduction chain and a Section 2 to 7 outline; RQ-driven empirical studies get the empirical

• sv-write-style-apply

  Applies a mined style profile to a manuscript without adding facts or changing technical content: section-by-section revision with diagnosis, revision and a per-paragraph log of rule candidates, or a report-only venue and field fit check for sv-review-loop round 2. Needs a dynamic writing skill from

• sv-write-style-mine

  Mines writing style from exemplar papers: dissects one paper into a Style Card, accumulates cards into venue, field or author profiles under a consensus threshold, finds exemplars by journal or author through public APIs, and borrows analysis methods as method cards. Describes moves, never content,

• sv-write-translate

  Rewrites a Chinese academic draft, or English with heavy Chinese-English residue, into submission-quality English by extracting each sentence's meaning first, restoring implicit logic, keeping terminology stable and fixing common residues, without inflating claims. Obeys the shared faithfulness rule

• sv-write

  Entry point for research writing: paper skeletons, Introductions and other sections, polishing and Chinese-to-English rewriting, critique without edits, de-AI editing of papers, reports, slides and letters, and mining or applying venue style. Routes to one sv-write sub-command and says which. Use wh

Similar plugins