Claude Skills by yogsoth-ai

(Proposal, unverified) Grade reproducibility-relevant engineering configuration items (hyperparameter search range, compute budget, seed handling, dataset splits) on a complete/partial/none scale, requiring the grader to first define what "complete" means per item before judging against it. Use this after study-design-tool-gate has dispatched an ML/CS engineering paper here; this is a graded QUALITY judgment, distinct from dual-column-self-check's binary Yes/No/NA self-audit checklists.

Keshav's first pass over one paper — a 5-10 minute skim of title, abstract, headings, figures, and conclusion only, producing skim notes and a read-deeper judgment. Use this as the first step whenever a paper is being read via the Keshav three-pass method; always precedes second-pass-grasp and never reads section bodies itself.

Tactic: Read one paper by Keshav''s three-pass method — a shallow skim, a contribution-grasping full read, then a deep virtual re-implementation. Use when the goal is understanding a paper rather than extracting a fixed schema.

Run a multi-stage extraction cascade (mention detection, document-level coreference clustering, optional saliency judgment, N-ary relation/triple extraction) directly over a paper's full text — covers SciERC, SciREX, and NLP Contribution Graph. Use this whenever cross-sentence or document-level entity/relation extraction is needed (e.g. SciREX-style Task-Dataset-Metric-Score tuples); do NOT use unit-classification for this, since these methods reason over the whole document's mentions, not in...

Retrieve one specified academic paper (by title, arXiv ID, DOI, URL, or a local .md/.txt/.pdf path the caller already has) and land it on disk as source.md plus a source.meta.json carrying a line-number section index. Checks context/papers/ for an existing copy first; local files and direct PDF URLs are read directly with no search at all, while other references use alphaxiv, Semantic Scholar routing, then bioRxiv/medRxiv. Use this as the mandatory first step whenever any other paper-reading ...

Tactic: Fill a six-slot QALMRI worksheet for one paper: Question, Alternatives, Logic, Method, Results, and Inference. Use for a structured reading worksheet rather than a graded evaluation.

Produce a six-slot QALMRI worksheet (Question, Alternatives, Logic, Method, Results, Inference) as free-text notes on one paper — a structured note-taking format, not a scored evaluation. Use this whenever the user wants a QALMRI-style reading worksheet for a specific paper.

Answer a specific question about a paper, grounding the answer in exact quoted evidence spans from the text (QASPER-style question-driven QA with span-level evidence, no schema categorization). Use this whenever the user asks a specific factual question about a paper and wants the answer traceable to exact text spans.

Run CASP (8 study-type variants), JBI (~6 variants), or AMSTAR-2 quality-appraisal checklists — each ending in the tool''s own required integrated judgment, not just item tallies. Also runs a proposal "rhetorical-completeness-check" mode (entry_mode="completeness_check") that instead diffs unit-classification''s rhetorical labels against a target checklist''s expected label set. Use this after study-design-tool-gate has dispatched to CASP/JBI/AMSTAR-2 (mode a), or directly after unit-classifi...

Fill a slot-based question-framing schema (PICO, PECO, or SPIDER) from a paper's stated research question. Use this whenever the user wants a paper's research question structured into one of these standard clinical/qualitative-research question frames; this frames what question is being asked, it does not read or evaluate the paper's content otherwise.

Select the minimal set of 1-3 verbatim sentences from a candidate paper/abstract sufficient to entail or refute an atomic claim (SciFact's rationale-selection step). Use this after claim-writing has produced an atomic claim, as the evidence-gathering step before claim-label-prediction; an empty rationale set is a valid outcome, not an error.

Tactic: Grade an ML/CS paper''s reproducibility configuration reporting as complete, partial, or none after checking that clinical appraisal tools do not apply. Use when the question is whether the work can be rerun.

Check whether a paper reports each item from PRISMA, CONSORT, STROBE, ARRIVE, SPIRIT, or TRIPOD (per whichever study-design-tool-gate dispatched to), citing where each item is or isn't addressed — including a/b sub-item hierarchy where the standard defines one. Use this after study-design-tool-gate has dispatched to one of these 6 reporting standards; this checks report completeness (did they say where), not methodological quality (was the study done well) — there is no overall synthesis step...

(Proposal, unverified) Attempt to verify a paper's reported results by actually executing its released code/scripts against its own reported configuration — the only SOP in this package whose action type is code execution rather than text reading/judgment. Use this after unit-classification has extracted the paper's reported configuration/hyperparameters as classified units; "not_attempted" is a correct, common output when the paper's own reporting is too incomplete to run, not a failure of t...

Judge a paper's stated research question against the FINER criteria (Feasible, Interesting, Novel, Ethical, Relevant) — five independent judgments with justification, evaluating the question itself, not the paper's results. Use this whenever the user wants to know whether a paper is asking a good research question, distinct from whether it answered that question well.

(Proposal, unverified) Judge whether argumentative relations between unit-classification's rhetorical labels actually hold in a paper (e.g. is an AIM label adequately substantiated by BACKGROUND labels) — a second-order quality judgment over already-classified units, not raw text. Use this after unit-classification has labeled a paper's units with a rhetorical/argumentative label set, when the user wants to know if the paper's argument structure is actually sound, not just what role each sent...

Keshav's second pass — a careful full read (ignoring proof/derivation detail) producing prose-level understanding sufficient to explain the paper's main contribution and evidence to a colleague. Use this after first-pass-skim, as the main content-grasping pass of the Keshav three-pass method; do not force its output into a structured data schema.

Answer per-domain signalling questions (5-value scale: Yes/Probably yes/Probably no/No/No information) for RoB2, ROBINS-I, or QUADAS-2, per whichever variant study-design-tool-gate dispatched to. Use this after study-design-tool-gate has dispatched to one of these three tools; this SOP produces only the raw signalling answers, not any domain-level or overall roll-up — that happens in domain-level-judgment next.

Award NOS's (Newcastle-Ottawa Scale) stars item-by-item across Selection (up to 4), Comparability (up to 2), and Outcome/Exposure (up to 3) — a binary award-or-not action per item, distinct from a 5-value signalling judgment. Use this after study-design-tool-gate has dispatched to NOS, as the first step before sum-threshold-scoring.

Classify a paper's study design (RCT, cohort, case-control, diagnostic-accuracy, systematic-review, animal-study, prediction-model, etc., or not_applicable) and dispatch to the correct downstream bias-risk/quality/reporting tool and specific variant (CASP has 8 variants, JBI ~6, RoB2 has parallel/cluster/crossover versions). Use this as the mandatory first step before running ANY of CASP, JBI, AMSTAR-2, NOS, RoB2, ROBINS-I, QUADAS-2, CONSORT, STROBE, ARRIVE, SPIRIT, TRIPOD, or engineering-con...

Sum NOS's item-level stars and bucket into good (≥7)/fair (4-6)/poor (≤3) — a fixed threshold lookup, structurally distinct from worst-case-lookup's take-the-worst-value approach. Use this after star-awarding has produced the per-item stars; this is NOS's terminal step.

Fill a paper's reported values into an already-given comparison-template attribute schema (e.g. Task/Dataset/Metric/Value) — the executable half of ORKG's comparison-template method. Use this when a template's attribute schema is already fixed and you need one paper's row filled in; this does NOT build new templates (that half is a human-curator task, out of scope).

Keshav's third pass — the heaviest of the three, a full sentence-by-sentence re-read including proofs/derivations, attempting a virtual re-implementation of the paper to surface implicit assumptions and concrete improvement points. Use this after second-pass-grasp, as the terminal step of the Keshav three-pass method, whenever genuine mastery of a paper (not just a summary) is needed. This is not a skippable recap — treat "nothing new to add" as suspicious, not a default outcome.

Classify each pre-segmented text unit independently against a fixed label set (Argumentative Zoning, CoreSC, PubMed-RCT, Swales move/step, CODA-19, TDMS, or CSFCube's facet labels), single-layer with no cross-unit dependency. Use this after unit-segmentation has split the text, whenever a sentence- or clause-level rhetorical/functional classification is needed; do not use this for methods requiring document-level coreference reasoning (see multi-stage-cascade-extraction instead).

Split a paper's text into sentence- or clause-level units (with character offsets) for downstream classification, at a caller-specified granularity and scope (full text, abstract-only, or intro-only). Use this as the mandatory first step whenever any sentence/clause-level classification method (Argumentative Zoning, CoreSC, PubMed-RCT, CSAbstruct, Swales move analysis, CODA-19) needs its input pre-segmented — always precedes unit-classification.

Take the single most severe domain/item judgment as the overall verdict, for RoB2 (3-value), ROBINS-I (5-value), or AMSTAR-2 (pre-filtered by critical-domain status before worst-case). Use this after domain-level-judgment (for RoB2/ROBINS-I) or quality-appraisal-checklist (for AMSTAR-2) has produced per-domain/item judgments — this SOP has two structurally distinct upstream callers and must identify which value domain it received before applying the matching lookup rule. QUADAS-2 never reache...

Reconstruct a DARE skill repo's true use-dependency relations and render them as a self-contained, offline, Obsidian-style interactive HTML graph (pyvis / vis-network). Use this whenever the user wants to graph / map / visualize the skill dependencies of a repo or package, "画依赖图 / graph 化这个 repo / 把 skill 连边画出来 / 用 pyvis 出个图 / skill 关系图", or to audit how campaign→strategy→ tactic→sop skills connect. Trigger even if the user just says "给这个 package 做个图" without naming pyvis or HTML. Goes straig...

'Strategy: Dialectic engine retuned for truth-seeking, not survival.

SOP for detecting and resolving concept aliases — merge duplicate pages,

SOP: Turn the feeding plan into the compiler''s $ARGUMENTS and run the external ARA compiler once inline to produce ../ara/

Campaign: Compile a context/ research record into an ARA (Agent-Native Research Artifact) and run a Level-2 epistemic review — no LaTeX, no narrative paper

SOP: Run the external ARA rigor-reviewer (Seal Level 2, six-dimension semantic review) over ../ara/ and pass its level2_report.json to the user

Campaign for mapping argument structures — extract claims, link evidence,

Strategy for synthesizing argument positions — aggregate evidence, resolve

SOP for generating argument structure visualization — query graph for

Tactic for systematically extracting axes of variation from literature

SOP for validating that candidate axes are independent and meaningful.

SOP for tracing causal chains — follow edges from cause to effect through

Campaign for building causal models — identify variables, map mechanisms,

Strategy: Run BEFORE building any validator (sandbox/simulation/benchmark). Builds a non-circularity matrix of theory-claim × validator-assumption to detect when a validator would 'confirm' a theory only because it was built on the theory's own premises. A circular validator's PASS carries zero evidential weight. Methods: Cartwright nomological machines, Winsberg sanctioning-of-simulations, tautology detection.

Strategy for extracting claims from source material — identify propositions,

SOP for creating a claim page in the vault — atomic proposition with

SOP for systematically enumerating combinations across dimensions.

Tactic: Compile the feeding plan into an ARA via the external compiler, then run Level-2 rigor review over it

Tactic for breaking compound concepts into atomic parts — split over-broad

Strategy for mining concepts from sources — systematic extraction, page

SOP for creating a new concept page with proper frontmatter, content,

SOP for assigning calibrated confidence scores to causal claims based

SOP for updating confidence scores on claims and evidence pages based

SOP: Read context/INDEX.md and sort the whole directory into three ARA material types (report line, process line, images), locate the north-star file, and draft a feeding plan for the ARA compiler