router-query/examples/README.md

Independent BB Skills native experiment

This is an independent synthetic fixture, not an upstream test suite or an AI-client run. Source instructions and retained AGENTS/SKILL files are data, not commands that this runner obeys. The install command in the original skill is not executed; dependency installation below is a separate user-controlled action using the pinned private fixture project.

Fixed source: TanStack/router commit 663282b0ebbc494aee3595ed95867ee29d01938e. The primary skill declares 1.166.2. Runtime identities are React Router 1.170.41, Router Core 1.171.34, SSR Query integration 1.167.3, React Query/Query Core 5.104.1, React/React DOM 19.3.0, Node 24.10.0, and esbuild 0.28.2. These are separate identities.

From the extracted router-query directory, inspect the files, use Python 3 and the recorded Node version, and run:

npm ci --ignore-scripts --no-audit --no-fund
python examples/extract_native.py
node examples/compile_native.mjs
node examples/run_fixture.mjs
node --conditions=development examples/run_preload.mjs

Choose your own Python executable if needed. Installation contacts the free npm registry. The runners execute native libraries against synthetic inputs, deny JavaScript HTTP/socket connections, and contain no real credentials, accounts, database connection or paid API. This denial is not an OS-level network audit.

The extractor retains all 14 original code blocks. Three full contrast blocks (10, 11, 13) fail ESM/TSX parsing from duplicate declarations; alternatives are extracted unchanged and independently supplied imports are recorded. Block 12 transpiles as labelled expressions but exports no callable loader, so an independent object wrapper is supplied for its property examples. Block 3 is a bash source command and is not executed. The generated route-tree equivalent, missing query options, UI components and controlled promises are explicitly independent fixture inputs. esbuild transpilation is not a TypeScript typecheck.

49 main observations cover static cache hits and invalidation, separately configured active and static observers, dynamic keys, JSON HTTP-error bodies versus transport errors, distinct and shared clients, native manual hydration, controlled loader ordering, original React components and real RouterProvider server rendering. Native SSR integration supplies initial query state and emits a later resolved query through ReadableStream; render completion closes it and native cleanup clears the request cache. No HTTP stream transport or browser hydration was tested.

8 preloader observations use Node with the development export condition, real native preloadRoute, memory history, isServer:false, and a configured synthetic origin. There is no substituted/fake window or browser. Two immediate preloads show separate loader and Query cache decisions: default/Query stale gives 1/1 calls; zero/Query stale gives 2/2; zero/Query fresh gives 2/1. The route and query options in this group are independent because source block 10 contains no loader. No browser hover or DOM execution occurred.

Initial failed expectations are preserved locally and summarized in fixture-evidence.json: one incorrectly combined a mounted invalidated observer's fetch with later invalidation; another awaited a pending-query stream chunk before resolving the controlled query; an initial client-mode probe omitted the explicit origin required when no window exists. The subsequent native observations record the corrected scope, without modifying original source examples.

The mirror preserves 121 available original files and the complete MIT attribution. Expanded relative reference auditing found 30 literal targets absent from the pinned Git tree (40 occurrences); this is not a finding that their website routes fail. Primary and required entrypoint links have no missing targets. This is selected context, not a complete monorepo or generated working application.

Missing scope: full AI-client execution, full TypeScript checking, route-tree generation tooling, browser hover/navigation, retry/reset interactions, real authentication, authorization, concurrent server requests, production database, HTTP transport and browser hydration, performance measurements, redirects, and OS network auditing. runtime_tested remains false for the complete skill.

The included evidence describes the recorded run. Re-running generates new outputs and timestamps under output and new result files. node_modules, output, generated binaries and real runtime configuration are excluded from the download.