Check named public GraphQL documents against the serving installation using the
existing compiler, linker and authorized snapshot interpreter. Keep listObjects
as a convenience over workspace-root queries, with no separate filter language.
Bound compilation/results, cache by content and serving plan, and fence cutovers.
Share graph discovery and reads between qx-workspace graph and the gardener's
workspace_graph tool. Local authoring shells discover the live endpoint through
the private control socket; configured public mounts also support direct clients.
Fix pending paginated inspections to wait for their confirmed window, and retain
semantic integration assertions instead of locking a server ticket's shape.
Replace the long prompt with a 384-word entry point and five focused guides.
Build immutable bundles with profiles, deep links, source hashes and word/link
checks. Keep the docs-maintenance process and feedback outside agent guidance.
Running agents use current guides while authored source and checker pins remain.
Validation: nix fmt; nix fmt -- --check; nix run .#lint; existing CLI checks;
protocol/client/coordinator build tests; complete Studio integration; 13 agent
unit tests. A real GPT-6.1 Sol medium/priority turn summarized two synthetic saved
comments and a document draft in 27.7s without source changes or cutover. Live
reads also succeeded after an ordinary check/activation in the retained fixture.
Keep query definition identity scoped to interfaces present in the generated
query schema. Exclude retention provenance from both interface revisions and
generic applications while preserving nominal identities, closed arguments,
operation contracts, documents, budgets and fragment dependencies. Advance the
query semantics version so prior compiler outputs cannot share the new identity.
Add a generic compiler regression covering unrelated imports, retained generic
sources and invalidation when an observed getter changes. This fixes Studio's
CLIENT_QUERY_MISMATCH when template contracts become editable local repositories.
Validation: protocol tests (296 total, 293 passed, 3 existing skips); nix fmt,
monorepo lint, final formatting check and diff check; camino-dev/qx-workspace
builds. All 17 Studio query descriptors match a rebuilt copy of the user's
workspace. Headless fresh-workspace canvas/library/Reader/reload check passes.
Repair the existing local qx-reader workbench through its ordinary check and
activation path, updating only its pinned protocol/checker CLI. Retain the
original toolchain GC root and record checker provenance. Source commit, runtime
and template pins, and database are preserved. Verify the selected candidate is
ready at epoch 2 and its canvas loads through the corrected Studio bundle without
creating application objects. Stop the independent smoke-test services.
Generate checked navigation query pairs from closed React contracts before
package checking. Retain per-type subscriptions under the logical
PresentationMount and resolve cold component metadata independently of code.
Keep scoped occurrence membership and idempotent local preparation in the
presentation library; copy directly through state contracts without reading
other PresentationMounts' source associations. Preserve normal transaction
creation, pending references, portal synchronization and detach.
Exercise automatic object refs in Reader queries, imported fragments and
structured/aggregate results. Add heterogeneous owner-index, live dependency,
cold metadata, browser ownership and latency coverage. Document measured
improvements, supported query limits and qualification evidence.
Implement scoped interface extents across QX, generated bindings, SDK,
Orch, Camino and browser queries. Preserve native projection, aggregation,
pagination and watch invalidation without synthetic persisted objects.
Validate exact query grants and scope freshness on execution and replay.
Add local PostgreSQL, transport, browser and code-generation coverage.
C1 implementation slice only: full-text indexes and Studio search adoption
(C2/C3) remain tracked in the implementation handoff. No publication or deployment.
Timothy J. Aveni