Skip to content

Build a portable customization toolkit

  • Create → load → use → inspect evidence → improve.
  • Three client views share one complete Markdown source. Pages adds selectable tabs; plain repository Markdown, no-JavaScript, and print show all steps.
  • Start and prerequisites · Client support and limits · Examples / safe recovery.
  • Main search is deliberately unfinished; later examples never silently complete it.

Progressive route

LabParticipant deliverableWorkspace
00 — StartVerified starter and actual client/buildAuthor
01 — InstructionsRepository/API/test rules and observed scopeAuthor
02 — Planning promptReusable planning-only invocationAuthor
03 — SkillOn-demand procedure, checklist, revisionAuthor
04 — PluginBuilt package, clean sibling, install evidenceAuthor → consumer
05 — MCP and updateFixture tool call, version 1.1.0 packageBoth, explicitly labeled
06 — RolesPlanner, implementer, independent reviewerConsumer
07 — Use toolkitApproved search change + added acceptance testConsumer
08 — Review and handoffSupported finding and reproducible continuationConsumer → author extension
09 — Spec-Kit (unavailable)Honest optional boundary, no fabricated generationReference only

Checkpoint entry

  • To join at Lab 04: read 00–02, author or explicitly import 01, 02, 03, then 04.
  • To join at Lab 07: complete/import author checkpoint 01–05; build the 1.1.0 package; create consumer; install once; configure MCP; load 06. Import 07's readiness checklist—not search.
  • Follow each dependency's preview/apply separately. Checkpoints never overwrite authored equivalents.
  • Every static-only or manual fallback is useful but must not be reported as native runtime completion.