| name | propose-exhibits |
| description | Generate a menu of candidate exhibits for the curator to choose from, grouped by hall, each with a one-line hook. Use when the curator asks "what next", wants options to pick from, wants to restock the backlog, or wants to see coverage across halls. |
Propose exhibits
Produce a menu the curator picks from. This skill only proposes - the curator
names the specific cases to build, then the add-exhibit skill builds each one.
Sources and filters
- Pull queued ideas from
.claude/memory/backlog.md.
- Exclude everything in
.claude/memory/rejected.md - never re-propose a
declined idea, and pre-filter anything that trips its reason categories
(predictable finale, can't reproduce, timing-only, CI-would-lie,
doesn't-happen-in-real-code).
- Apply the curation bar from
CLAUDE.md: one broken mental model, reject
predictable finales, prefer accessible + axiomatic, prefer silent-wrongness /
money stakes.
- Balance against
.claude/memory/archetypes.md - spread across the 7
archetypes, don't stack one.
Format (grouped by hall)
Group candidates under their hall (the canonical hall list is
.claude/memory/halls.md). Mark planned halls (a candidate there opens the
hall). One line per candidate:
slug - the twist (the mechanic, not just the restated title)
The description is the "wait, WHAT?" - the hook that makes it exhibit-worthy,
not a full explanation. Keep the whole menu scannable.
After proposing
Stop and let the curator select. Do not start building. When cases are named,
switch to add-exhibit for each. Move chosen ideas out of backlog.md as they
ship; add declined ones to rejected.md with a reason.
