CL-6529: Fix false inference-catalog claim, posix-tools.md fiction, fold structural checks into check

docs/model-seeding.md

@@ -62,8 +62,15 @@ model/provider name in the codebase is a literal string someone typed in.
 Read from the upstream Interchange checkout's `packages/` directory (upstream,
 not vendored into this repo): `inference-discovery`,
 `inference-discovery-{anthropic,google-genai,openai}`, and
-`inference-testing`. There is no `@intx/inference-catalog` package — it does
-not exist upstream or on npm (confirmed via `npm view`).
+`inference-testing`. These are separate from `@intx/inference-catalog`,
+which does exist and already ships in this repo: pinned in `package.json`
+at `0.3.0` and imported directly (not vendored — see `VENDORED.md`) by
+`packages/inference-catalog/src/offering-capabilities.ts` for
+`catalogProviders`, the pinned catalog data that `seedCatalog` reads
+through today (see "How workbench seeds models and providers today",
+above). It is a baked catalog of prior probe results, not a discovery
+service — it has no relation to `inference-discovery`/`inference-testing`
+beyond the name.
 
 `@intx/inference-discovery` is **not** a live model/provider discovery
 service. Per its own README, it is the runtime for Interchange's internal

docs/posix-tools.md

@@ -1,14 +1,24 @@
-# Posix tools (design note)
+# Posix tools (design note — not implemented)
 
-`@corbits/posix-tools` runs only on a workbench placed on isolated capacity
+Nothing under this name ships in workbench today. There is no
+`@corbits/posix-tools` package anywhere in this tree. Upstream Interchange
+does publish a real package in this space, `@intx/tools-posix` — its
+`sidecar-bundle:run_shell` tool already appears in this repo's fixtures
+(`packages/mocks/src/ollama/scenarios.ts`,
+`vendor/intx/hub-sessions/src/event-collector.test.ts`) — but workbench
+does not depend on it and has built no integration against it. Treat
+everything below as a proposal for that integration, not a description of
+current behavior.
+
+The design: run only on a workbench placed on isolated capacity
 (`sidecarPlacement` enabled, backed by a real `SidecarProvisioner`) — the
 dedicated container is the sandbox boundary, so there is no in-process
-sandboxing to build. Tool set: `shell`, `read`, `write`, `glob`, `grep`.
-Every write-capable tool (`shell`, `write`) declares `approval: "ask"`, so
-each call surfaces through the existing human-approval path before it runs
-— no tool in this package ever gets a standing auto-approve. The sandbox
-policy is a named, auditable, per-workbench setting (alongside
-`sidecarPlacement`), defaulting closed until an operator opts a workbench
-in — never a single all-or-nothing flag. This mirrors the same pattern
-already used for exclusive sidecar placement itself: capability is inert
-until named, scoped, and explicitly turned on.
+sandboxing to build. Every write-capable tool it exposes (shell, write)
+would declare `approval: "ask"`, so each call surfaces through the
+existing human-approval path before it runs — no tool in this package
+would ever get a standing auto-approve. The sandbox policy would be a
+named, auditable, per-workbench setting (alongside `sidecarPlacement`),
+defaulting closed until an operator opts a workbench in — never a single
+all-or-nothing flag. This mirrors the pattern already used for exclusive
+sidecar placement itself: capability is inert until named, scoped, and
+explicitly turned on.

package.json

@@ -11,7 +11,8 @@
     "workflows/*"
   ],
   "scripts": {
-    "check": "bun run typecheck && bun run lint && bun run test",
+    "check": "bun run typecheck && bun run lint && bun run test && bun run check:structural",
+    "check:all": "bun run check && bun run check:packages",
     "typecheck": "bun run scripts/run-all.ts typecheck",
     "lint": "prettier --check . && eslint .",
     "format": "prettier --write .",
@@ -23,6 +24,7 @@
     "setup:memory": "bun run scripts/setup-memory.ts",
     "seed": "bun packages/cli/src/index.ts seed",
     "reset": "bun packages/cli/src/index.ts reset",
+    "check:structural": "bun test scripts/checks/test && bun run check:deletion && bun run check:killdates && bun run check:licenses && bun run check:no-product-tenancy && bun run check:browser-safe-subpaths && bun run check:web-utilities && bun run check:tailwind-source && bun run check:ui-vocabulary && bun run check:react-ui-drift && bun run check:react-ui-pin && bun run check:tool-package-pins",
     "check:deletion": "bun run scripts/checks/deletion.ts",
     "check:killdates": "bun run scripts/checks/killdates.ts",
     "check:packages": "bun run scripts/checks/packages.ts",