Agentic skills for Claude Code and Codex, built from published social-science methods sources. Covers experimental design, computational text analysis, manuscript QA, and transparent reporting.
Command
Build or audit a literature review for the task below. Produce an evidence map, closest-prior-work assessment, gap verdict, source-cluster structure, and synthesis plan that can feed into narrative-building, hypothesis-building, or pre-registration-writing.
Command
Read a single model's own uncertainty from token log-probabilities and check whether that confidence is calibrated. Collect logprobs (OpenAI/vLLM/Ollama) at temperature 0 with a fixed seed, aggregate multi-token labels into one per-item confidence, map confidence (and the margin to the runner-up) onto triage tiers…
Command
Apply methods reporting expertise to the task below. Run through the reporting checklist covering CONSORT standards, JARS pre-registration elements, DA-RT transparency requirements, and open science infrastructure as relevant.
Command
Run the model-committee skill with the Fable 5.1 chair. Members are unchanged — GPT-5.6 "Sol" and Claude Opus 5 — and only the post-round-3 step moves: Fable validates schemas, aggregates the weighted scores, applies the precommitted tie rule, and writes the decision without voting its own prior a third time. Delegate…
Command
Run the model-committee skill with the GPT-5.6 "Sol" chair. Two things change from the default, not one: Sol chairs instead of Opus 5, and the GPT member drops from Sol to gpt-5.6-terra so the chair is not also a member. Members are therefore GPT-5.6 "Terra" and Claude Opus 5; Sol aggregates the scores, applies the…
Command
Run the model-committee skill with the Claude Opus 5 chair (the default). Members are GPT-5.6 "Sol" and Claude Opus 5.
Command
Run a contested labeling, scoring, or term-discovery task through a diverse panel of language models as independent coders, then read their (dis)agreement: assemble jurors from different training families to decorrelate errors, keep votes independent (no deliberation), apply a pre-stated k-of-N consensus rule with…
Command
Apply scientific narrative expertise to the task below. Cover introduction logic, literature framing, the Why-to-If-Then funnel, cumulative framing, and multi-experiment coherence as relevant.
Command
Run the orchestrate skill with --lead opus, forcing Opus-lead mode rather than detecting the lead from the session's model. Claude Opus 5 leads at medium reasoning effort by default (high only if the session will be dominated by direct hard reasoning). You are the same model as the deep-reasoner, so reason compact…
Command
Act as the multi-model orchestrator. Pick the lead first: read the model line in your own context ("You are powered by the model named …") and run Fable-lead mode on Fable 5.1 or Opus-lead mode on Claude Opus 5. On Sonnet or an unrecognized model, say which model you detected in one line and run Opus-lead mode without…
Command
Alias for /oss:paper-review-lite --codex. Run the paper-review-lite skill in its cross-model mode: Claude and Codex (GPT-5.6 "Sol" at xhigh effort) independently apply the nine review dimensions, each cross-checks the other's findings, and every retained Critical or Recommended issue carries a confidence label.
Command
Run a Critical-Reviewer-style pre-submission audit of the current paper using parallel sub-agents inside Claude Code. Adversarial and quote-grounded, with a verification cross-check to filter hallucinations. Covers content and argument, numerical consistency, references and DOIs, writing quality, figures and…
Command
Typeset the draft below as a house-style LaTeX paper and build the PDF. Detect the input format and convert the body with pandoc, wrap it in the EB Garamond template under assets/, and build with latexmk via scripts/formatpaper.py. Then do the house-specific finishing by hand: \figcap title+note captions, [H] floats…
Command
Run the vlm-ocr skill in its clean phase: LLM and rule-based correction of raw OCR text, quality diagnostics, multilingual handling, corpus-level QA, and span-level provenance.
Command
Apply pre-analysis plan expertise to the task below. Cover PAP structure, registry selection, analytical strategy specification, confirmatory vs. exploratory distinctions, and deviation documentation as relevant.
Command
Activate the standalone presubmit Python CLI — a 30+ stage adversarial peer-review pipeline (Red Team, Blue Team, verification cascade, legal pass, copyedit, Writer Mode) that calls the Anthropic API directly and writes a consolidated review report to disk. Walks first-time users through install (clone + venv + pip)…
Command
Apply Qualtrics live-survey operations expertise to the task below. In the default operations mode, cover when to use live-survey operations versus design/build, non-negotiables (backups, read-back verification, the version list as proof, one writer at a time, asserting untouched fields are unchanged), publish gating…
Command
Organize and format the author's response to the referee reports below. Extract every distinct point with severity and type, order the revision by dependency, flag defensible pushbacks as questions for the author, and build the response letter as a numbered comment → response → location table with the substantive…
Command
Scaffold or audit a social-science replication package at a target directory, and audit the manuscript and its archived research objects against FAIR principles.
Command
Interview the researcher in rounds about the idea, design, or draft below until no decision is left silently assumed. Infer the stage (idea, design, defend) from what they bring, or take it from the arguments; --plan runs the generic plan interview. Ask the whole frontier each round, numbered, each question in plain…
Command
Scaffold a new research project repository, or audit an existing one, organized around its source library. The source library (sources/) is the spine: original PDFs in og/ (gitignored), LLM-readable Markdown in md/ (tracked), a drop zone in unprocessed/, and a references.bib keyed to it. From that spine the skill…
Command
Plan the research project below as a decision map that outlives any single session — adapted for research from Matt Pocock's wayfinder. If no map exists, chart one: interview the researcher until the destination (a defensible, pre-registerable design) is concrete, write planning/map.md plus typed decision tickets…
Command
Spawn one or more full Claude Code peer sessions (real sessions in their own terminal panes and git worktrees, not subagents), each on a directed task. Detect the environment first (herdr if HERDRENV is set, else tmux, else a native claude --bg background agent), create a worktree on branch spawn/ per task, write the…
Command
Apply survey data audit expertise to the task below. Cover when to use the audit, required and optional inputs, the precedence of registered definitions over defaults, structural completeness, the hard-signal automation composite, the report-only descriptive battery (bot/AI screening, fraud and reCAPTCHA scores…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: