From epistemic-skills
Enforces a research discipline layer for coding agents doing evals, benchmarks, or empirical comparisons. Describes activation, stage skills, and research gates to ensure reproducibility.
How this skill is triggered — by the user, by Claude, or both
Slash command
/epistemic-skills:using-epistemicThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Epistemic is a research-discipline layer for coding agents. It exists to stop
Epistemic is a research-discipline layer for coding agents. It exists to stop empirical work from turning into cherry-picked runs, stale baselines, or headline claims with no paper trail.
If the task involves empirical research, evals, benchmarks, model comparisons,
ablation studies, "is X better than Y?", or publishing a result, invoke the
epistemic skill before answering or acting.
If a stage-specific skill applies, invoke it next:
research-questionpreregistrationbaseline-reproductionexperiment-executionstatistical-rigorfalsification-reviewsurprise-triagekill-or-shipverification-before-publicationDo not skip ahead because a run looks cheap, obvious, or exploratory. A cheap unregistered run can still contaminate the evidence chain.
Follow explicit user and repository instructions first. Epistemic skills define how to do research work, but they do not override direct user constraints, repository policy, or safety requirements.
Treat a repo as Epistemic-active when it contains any of:
HYPOTHESES.mdexperiments/RESULTS.mdIf none exists but the user asks to start empirical work, use
research-question to create the first falsifiable hypothesis before running
benchmarks, evals, training, or judge-backed scripts.
In the OMP agent, several gates are runtime-enforced. In portable harnesses such as Claude Code, you must enforce them yourself:
experiments/<id>/prereg.md.OVERRIDES.md entry explaining why.smokes/ number in README.md, RESULTS.md, a paper, or a claim summary.surprise-triage when a result diverges by more than 15%.kill-or-ship when cost exceeds 1.5x cap or a hypothesis goes stale.Overrides are allowed only when recorded with a concrete reason. If the reason would embarrass you in a methods section, do not override.
Claude Code should use the Skill tool for full skill content. Other harnesses
may expose skills differently; use their native skill mechanism when available.
The method is portable even when runtime gates, dashboards, or OMP-specific
widgets are not.
When Epistemic applies, start with a short announcement:
Using the epistemic mechanism.
Then load the umbrella epistemic skill and the current stage skill before
continuing. Ask one clarifying question at a time when defining a hypothesis.
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.
npx claudepluginhub atomicstrata/epistemic --plugin epistemic-skills