Skip to content

Teach agents to compose Constructive applications with App Kit - #205

Draft
yyyyaaa wants to merge 4 commits into
mainfrom
feat/app-kit-skills
Draft

Teach agents to compose Constructive applications with App Kit#205
yyyyaaa wants to merge 4 commits into
mainfrom
feat/app-kit-skills

Conversation

@yyyyaaa

@yyyyaaa yyyyaaa commented Aug 3, 2026

Copy link
Copy Markdown
Contributor

Summary

  • extend constructive-blocks with an intent-first application-composition lane and a separate optional platform-capability lane
  • route by user intent, data shape, presentation geometry, and workflow type instead of department labels
  • keep Skills procedural and progressively disclosed while treating Blocks documentation and registry metadata as the exact API authority
  • add directly linked composition, resource, view, action/workflow, verification, and Event Studio guidance plus small cross-domain recipes
  • add validated --family and --capability catalog queries and deterministic brief-to-root fixtures that reject defaulting to Sheets, Console Kit, review queues, or Event Studio

Stack relationship

This draft is pinned to Blocks commit 1a72e5d95f7ce4a243cd4536ed78c638708d538c. It deliberately keeps publicRegistryReady: false and must remain a stacked draft until the Blocks PR merges, public Pages registry URLs are verified, the branch is refreshed from latest main, and generated attestations/catalogs are pinned to the merged SHA.

Verification

  • 35/35 Node tests
  • skill and reference validators
  • portable and live source checks
  • family/capability catalog queries
  • generated catalog and sync checks

Fresh Terra-high forward validation against the exact pinned Blocks and Skills
heads reached discovery in 139,437 ms, root installation in 221,689 ms, first
tenant data in 785,010 ms, first successful action in 826,083 ms, and negative
authorization proof in 843,233 ms. The agent recorded a production-build
milestone at 950,393 ms, but its immutable evidence did not independently prove
the build, so that timing is not counted. It did not reach candidacy: the
generated consumer's rendered-session
authorization proof, relation persistence/cleanup, overlay dismissal, and edit
coverage were incomplete, so the run is excluded and time-to-accepted-app
remains unknown.
Those generic failures have been folded into the persisted next-run prompt and
observer checks without duplicating Blocks API documentation. Per the session
delivery constraint, this PR stays a draft; the post-merge public-readiness step
is still required afterward.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant