episteme

Run one bounded episteme Event end-to-end — branch, implement, verify, PR, merge, sync, doctor — with the repo's release and hook machinery handled correctly. Use when starting or closing a unit of work in the episteme repo (or any repo adopting its flow). Triggers on "new event", "이벤트 시작", "PR까지", "merge하고 sync", "release", "릴리즈", "브랜치 파서 작업".

Author the episteme gate artifact (.episteme/reasoning-surface.json) correctly on FIRST contact — schema, lens order, honest values, strict-vs-advisory, and when to escalate to epistemic-interrogation instead. Use the moment a PreToolUse hook reports a missing or stale Reasoning Surface, or BEFORE starting a session's first high-impact operation. Triggers on "reasoning surface", "Episteme Strict Mode", "surface stale", "blocked by episteme", "게이트 막힘", "서피스 갱신".

Devil's Advocate stress-testing for code, architecture, PRs, and decisions. Surfaces hidden flaws through structured adversarial analysis with metacognitive depth. Use for high-stakes review, stress-testing choices, or when the user wants problems found deliberately. NOT for routine code review (use /code-review). Triggers on "스트레스 테스트", "stress test", "devil's advocate", "반론", "이거 괜찮아", "문제 없을까", "깊은 리뷰", "critical review", "adversarial".

Senior-researcher interrogation of a load-bearing decision or conclusion — decompose into claims, verify the load-bearing ones in a fresh context with external evidence, argue the opposition, name the weakest link, pre-commit disconfirmation, record the verdict artifact the episteme gate reads.

Design and run unattended or long-horizon loops (overnight hardening, drain queues, iterative sweeps) with a written invariant the harness re-reads every iteration — no manual re-pasting, trustworthy metrics, explicit stop conditions, and correct model routing. Replaces bounded-loop-runner. Use for "overnight", "unattended", "keep iterating until", "밤새 돌려", "루프 돌려", ralph-loop or /loop driven work.

Prevents premature execution on ambiguous requests. Analyzes request clarity using 5W1H decomposition, surfaces hidden assumptions, and generates structured clarifying questions before work begins. Use at the start of any non-trivial task, or when a request could be interpreted multiple ways. Triggers on "뭘 원하는건지", "요구사항 정리", "clarify", "what exactly", "scope", "requirements", "정확히 뭘", "before we start".

Update progress and next-step docs so the next agent or session can resume with minimal context loss.

Prospective failure analysis using Gary Klein's pre-mortem technique (adapted here as swing-mortem). Assumes complete failure, works backward to identify risks, leading indicators, and circuit breakers. Counters optimism bias by forcing systematic exploration of failure modes before they materialize. Use for project plans, architecture decisions, technology adoption, business strategy, or feature launches. Triggers on "리스크", "위험", "실패하면", "swing-mortem", "뭐가 잘못될 수 있어", "risk", "what could go wrong", "걱정되는 점", "failure modes", "리스크 분석", "위험 분석".

Bootstrap a repository with the standard episteme scaffold and verify the core memory files exist.

Private Claude-only lab skill for iteratively improving a copied skill candidate with explicit evaluation goals and manual promotion.

Deep research with cross-verification and source tiering. Use when investigating technologies, comparing tools, fact-checking claims, evaluating architectures, or any task requiring verified information. Triggers on "조사해줘", "리서치", "research", "investigate", "fact-check", "비교 분석", "검증해줘".