Help us improve
Share bugs, ideas, or general feedback.
From epistemic-skills
Enforces a rigorous research pipeline for empirical/ML claims: hypothesis, preregistration, baseline reproduction, experiment execution, statistical rigor, falsification review, and kill-or-ship decision gates.
npx claudepluginhub atomicstrata/epistemic --plugin epistemic-skillsHow this skill is triggered — by the user, by Claude, or both
Slash command
/epistemic-skills:epistemicThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
A coding agent will happily cherry-pick a number that confirms what you hoped.
Provides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.
Explores codebases via GitNexus: discover repos, query execution flows, trace processes, inspect symbol callers/callees, and review architecture.
Share bugs, ideas, or general feedback.
A coding agent will happily cherry-pick a number that confirms what you hoped. Epistemic is the discipline that stops that: you commit to what would disprove your idea before you run anything, reproduce the baseline you claim to beat, and treat every result as guilty until it survives attack. Killing a bad idea fast is a win, not a failure (target ~5 kills per ship).
Announce at start: "Using the epistemic mechanism."
This skill is the orchestrator. Each stage below is its own skill — invoke them in order; don't skip ahead.
experiments/<id>/prereg.md.smokes/ only.In the epistemic agent these are hard-blocking. Here YOU enforce them:
prereg.md exists. Non-zero spend before prereg = protocol breach.smokes/ number in a headline file.Overrides are allowed but must be recorded in OVERRIDES.md with a reason
(≥50 chars). If you catch yourself rationalizing ("just one quick run first"),
that's the breach — stop and pre-register.
Don't serialize on one hypothesis. Run several lanes concurrently — each its own
experiments/<id>/ with its own prereg — and kill the losers fast. Fan out
variants (a sweep) as parallel subagents where they don't share a judge/baseline
lock.
HYPOTHESES.md # one ## Hypothesis: <id> block per lane
BASELINES.md # reproduced baselines (name, score, judge, version)
RESULTS.md # confirmed, verified results only
OVERRIDES.md # gate overrides, each with a reason
experiments/<id>/
prereg.md judge.lock smokes/ # provisional runs live in smokes/