From leopold
Phase 1 of Leopold. A structured debate that captures the mission and your decision-making, then writes the brief (MISSION, CHARTER, GUARDRAILS, PLAN) that the autonomous run executes.
How this skill is triggered — by the user, by Claude, or both
Slash command
/leopold:leopold-briefThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
You are running Phase 1 of Leopold. Your job is to turn the user's intent into a
You are running Phase 1 of Leopold. Your job is to turn the user's intent into a durable brief the autonomous run can execute without you. This is a debate, not a form. Push back, surface tradeoffs, and name what is missing.
The brief is the contract. The autonomous run never invents intent; it executes what you write here. The quality of the run is capped by the quality of this brief, so do it well.
Quietly check for a Leopold update; if one is available, tell the user. If
~/.leopold/auto-update exists, update now (the brief is a safe point).
UP="$(bash ~/.claude/leopold/scripts/leopold-update-check.sh 2>/dev/null || true)"
if [ -n "$UP" ]; then
echo "$UP"
[ -f ~/.leopold/auto-update ] && bash ~/.claude/leopold/scripts/leopold-update.sh || echo "Update with: make update (or /leopold-update)"
fi
Run: mkdir -p .leopold
If .leopold/MISSION.md already exists, read all existing artifacts and offer to
revise rather than overwrite.
Through conversation (use AskUserQuestion for real forks), establish:
Challenge vague goals. If the user cannot state a definition of done, the mission is not ready; help them find it.
Elicit how the user makes decisions, so the run can decide as they would:
Ask for concrete examples. "I hate clever abstractions" is more useful as "prefer a 10-line obvious function over a reusable helper unless it is used 3+ times."
Default posture (the recommended one): git commit, push, and publish stay LOCKED; the run stages and reports, the human commits. Confirm or adjust:
Decompose the mission into an ordered, checkbox backlog in PLAN.md. Each item
should be independently completable and verifiable. Order by dependency, then by
value. Keep items small enough that one is a reasonable unit of autonomous work.
If gstack is installed (check for ~/.claude/skills/gstack/, or whether the
/spec skill exists), offer to sharpen the plan before writing it. gstack's
planning skills are excellent here:
/autoplan or /plan-eng-review./plan-ceo-review./spec.Run them in spawned mode (prefix any gstack bash with OPENCLAW_SESSION=1) so
they auto-decide and report instead of prompting, then fold the result back into
PLAN.md. If gstack is not installed, skip this step silently; the plan you
already built is enough. gstack is optional and separate: https://github.com/garrytan/gstack
Copy the templates from the Leopold install (~/.claude/leopold/templates/) and
fill them in, writing to:
.leopold/MISSION.md.leopold/CHARTER.md.leopold/GUARDRAILS.md.leopold/PLAN.mdThen create an empty .leopold/DECISIONS.md with a header.
Summarize the brief to the user in a few lines: mission, the 3 sharpest charter rules, the guardrail posture, and the first 3 plan items. Ask if it reflects how they actually think. Iterate until they confirm.
End by telling them: when ready, run /leopold-run to hand over the seat.
npx claudepluginhub jonhvmp/leopold --plugin leopoldProvides 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.