From just-claude-things
Use as the last planning gate before scaffolding — stress-tests a completed plan (requirements + tech-design + tickets) for failure modes via a blind dual-draft pre-mortem, then emits a failure register with mitigations or accepted risks. Requires complete tech-design and issues docs. Triggers include "pre-mortem", "premortem", "how could this fail", "what could go wrong", "stress-test the plan", "risk the plan before we build".
How this skill is triggered — by the user, by Claude, or both
Slash command
/just-claude-things:pre-mortemThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
<what-to-do>
Run a Klein-style pre-mortem on a completed plan, before any resource is committed to building it. Assume the project has already failed badly; work backward to find why.
The defining mechanic is the blind dual-draft in PREMORTEM-LOOP.md: Claude
writes its own failure list to a sealed file before asking the user anything, the
user brain-dumps theirs freeform, and the user reveals first. This protects the
independence that makes a pre-mortem work — voicing a mode first anchors the other
party and collapses dissent (a social Silent Fallback). Do not short-circuit it.
Capture every mode in docs/pre-mortem/<slug>.md as it surfaces. Each mode ends as a
proposed upstream edit (applied only on approval) or an explicitly accepted risk. The
register never silently rewrites the design — it preserves the mode → mitigation chain.
docs/tech-design/<slug>.md at status: complete — Operational lens.docs/issues/<slug>.md at status: complete — Delivery lens.docs/requirements/<slug>.md at status: complete — Outcome lens (goal/metric).
See PREREQ-CHECK.md. If any is missing or in-progress, stop and name the upstream
skill (tech-design-grill, to-issues, or requirements-grill).CONTEXT.md, docs/adr/*.md, recent commits,
AGENTS.md. See PREREQ-CHECK.md.docs/pre-mortem/<slug>.md — the failure register. See FAILURE-REGISTER-FORMAT.md.requirements-grill → tech-design-grill → to-issues → [pre-mortem] → scaffold-with-guardrails
Pre-mortem is the last planning gate, not the last step. The plan must be whole but execution must not have started — scaffolding commits resources. Running the pre-mortem after scaffold would be a Paper Tiger: it would emit a risk doc after lock-in, unable to prevent the failure it exists to prevent.
PREREQ-CHECK.md. Refuse to start without all three complete docs.CONTEXT.md) for failure
modes so they land in domain language.Follow PREMORTEM-LOOP.md exactly, in order: frame → blind dual-draft (Claude commits
first, silently) → human reveals first → merge + dedupe → lens gap-check → score →
resolve → clean up the seal. Run LENS-CHECK.md for coverage after generation,
never before — running it first would anchor the user.
Capture inline. Update the register after each phase. Do not batch.
Run NEG-AUDIT.md. Beyond the standard violation guide, the anti-theater gates are
hard:
novel_mode_count >= 1; a zero gap is suspicious, not passing.plan_changed is true or ≥1 risk is explicitly accepted; a
no-change run must be justified, not auto-passed.mundane_mode_count >= 2.Flip status: complete only if the neg-audit passes.
PREMORTEM-LOOP.mdLENS-CHECK.mdFAILURE-REGISTER-FORMAT.mdPREREQ-CHECK.mdFRONTMATTER-FORMAT.mdNEG-AUDIT.mdCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub smalls257/just-claude-things --plugin just-claude-things