@artooi/ag-ui-web-component 0.14.1 → 0.16.0

package/CHANGELOG.md

@@ -7,6 +7,108 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.16.0] — 2026-08-09
+
+### Added
+
+- **A stale-page guard on frontend tool calls.** A round's context records the
+  page it describes; if the page moves before the agent's tool call arrives, the
+  call is refused with a result telling the agent to call `read_page` and retry,
+  instead of running the handler. Most stale calls would simply miss and report
+  a failure — the case this prevents is the other one, where a same-named
+  control on the *new* page matches and the agent silently acts on the wrong
+  page. `read_page` itself and tools marked `x-navigates` are exempt, and the
+  guard is inert unless a `getPageMap` provider is set.
+- **New `UiStrings` keys `runInterrupted` and `pageMoved`**, both overridable
+  like every other string.
+
+### Changed
+
+- **Toolchain majors: Vitest 3 → 4 and TypeScript 5.9 → 7**, plus `marked`
+  18.0.9, Biome 2.5.7, `@types/node` 26.1.2. All development dependencies — the
+  emitted `.d.ts` files are **byte-identical** to the 5.9 output (verified by
+  building both and diffing; only the source maps move), so consumers see no
+  change.
+
+  ⚠ **Vitest 4 takes a provider *instance*, not the string `"playwright"`.**
+  The provider moved to its own package (`@vitest/browser-playwright`) and, with
+  v8 coverage, the old string form is a hard error rather than a deprecation.
+
+  ⚠ **TypeScript 7 requires `rootDir` explicitly** (TS5011) instead of
+  inferring it from the common source directory. Set to the value 5.x inferred,
+  so the published layout is unchanged.
+
+### Fixed
+
+- **A run interrupted by navigation no longer vanishes silently.** If the page
+  navigates or reloads while a run is in flight — routine in an MPA — the
+  element is destroyed with it, and on the next mount the transcript replayed
+  with the answer simply missing and no indication anything had gone wrong. The
+  element now reports it as an inline notice. Detected from the shape of the
+  transcript (`send()` persists the user turn *before* starting the run, so a
+  history ending on that turn means nothing came back), which is why it needs no
+  `ClientConversationStore` change and no `pagehide` listener — neither of which
+  would fire on a crash or a force-quit anyway.
+
+  It is deliberately **a notice, not a resume**: AG-UI has no
+  resume-an-aborted-run primitive, so re-sending the accumulated messages is
+  semantically a new run and would re-execute any server-side tool the agent had
+  already performed. The agent-initiated case is unaffected — a navigating tool
+  still checkpoints and resumes exactly as before.
+
+- **Six branches that were never actually covered.** Vitest 4's v8 provider
+  remaps coverage more precisely, and the 100% gate stopped being satisfiable —
+  not because anything regressed, but because v3 had been crediting six
+  branches and three callbacks that no test reached. Each is now genuinely
+  tested: the paperclip button opening the file picker, the built-in
+  transcription handler (every prior voice test supplied its own), a page-action
+  tool resolving through `resolvePageTarget`, a non-`Enter` keystroke in a
+  question card, a submit click with no answer, a non-string `error` in a
+  transcription error body, and a restored history message with an unrecognised
+  role.
+
+  ⭐ **One was a flaw in the test harness, not a missing test.** `makeFakeAgent`
+  ended a clean run by calling `onRunFinalized` alone, so the client's
+  `RUN_FINISHED` path could only ever be reached through `emit.interrupt()` —
+  the ordinary success outcome every real run carries was never exercised. The
+  fake now emits both events, in the order a real agent does.
+
+  `route_map`'s unreachable guard was restructured away rather than tested: its
+  regex capture group is mandatory, so the `undefined` case existed only to
+  satisfy `noUncheckedIndexedAccess` and no test could ever have reached it.
+
+## [0.15.0] — 2026-08-08
+
+### Changed
+
+- **Sanitisation is now tested in a real browser.** `vitest.config.ts` defines
+  two projects: **happy-dom** for the bulk of the suite, and **Chromium**
+  (Playwright) for the tests whose subject is sanitisation. Coverage stays
+  unified at 100% across both.
+
+  ⚠ **A correctness requirement, not an optimisation.** DOMPurify 3.4.8+
+  silently stops sanitising under happy-dom — `<script>` and `<img>` pass
+  straight through, and ordinary markdown loses its `<p>` wrapper. A
+  happy-dom-only suite can therefore go green while this component ships no
+  sanitisation at all, which is the one failure it must never ship.
+
+  ⭐ **The experiment settles what the pin never could**: dompurify 3.4.13
+  sanitises correctly in Chromium. The defect is happy-dom's DOM emulation, not
+  a DOMPurify regression — so **consumers were never exposed**, and the risk was
+  confined to the test environment the whole time.
+
+  The browser run also *removes* a workaround instead of carrying it: happy-dom
+  eagerly **executed** inline `<script>` while DOMPurify parsed into its scratch
+  document, so the suite had to stub `alert`. A real browser parses into an
+  inert context.
+
+### Security
+
+- **`dompurify` is a normal caret range again** (`^3.4.13`), lifting the
+  exact-version pin — and with it the **five advisories** that pin was holding
+  open (three LOW, two MEDIUM). The browser project is what made that safe, and
+  is the standing acceptance check.
+
 ## [0.14.1] — 2026-08-07
 
 ### Security
@@ -646,7 +748,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Notes
 - First release — exercising the automated npm OIDC publish pipeline end-to-end.
 
-[Unreleased]: https://github.com/Artui/ag-ui-web-component/compare/v0.14.1...HEAD
+[Unreleased]: https://github.com/Artui/ag-ui-web-component/compare/v0.16.0...HEAD
+[0.16.0]: https://github.com/Artui/ag-ui-web-component/compare/v0.15.0...v0.16.0
+[0.15.0]: https://github.com/Artui/ag-ui-web-component/compare/v0.14.1...v0.15.0
 [0.14.1]: https://github.com/Artui/ag-ui-web-component/compare/v0.14.0...v0.14.1
 [0.14.0]: https://github.com/Artui/ag-ui-web-component/compare/v0.13.0...v0.14.0
 [0.13.0]: https://github.com/Artui/ag-ui-web-component/compare/v0.12.0...v0.13.0

package/README.md

@@ -648,13 +648,24 @@ chat.routeMap = [
 
 **`getPageMap(): PageMap`** — a per-run provider returning the current page's compact actionable
 surface (field names/types/labels, button labels+handles — *not* values). It is auto-injected into
-each run's `context` as a `page_map` entry (toggle with `autoInjectPageMap`). Recomputed every run
-so it reflects the page the agent is currently looking at:
+each run's `context` as a `page_map` entry (toggle with `autoInjectPageMap`):
 
 ```js
 chat.getPageMap = () => ({ fields: introspectForm(), buttons: visibleButtons() });
 ```
 
+It is recomputed at the top of **every tool round**, not once per `send()` — so after the agent
+acts, the next round already sees the resulting page. Within a round the agent can pull a fresh
+view at any time with the built-in `read_page` tool, which is registered whenever this provider is
+set.
+
+That leaves one window: the page can move *after* a round's context was built but *before* the
+agent's tool call arrives — the user clicks a link, or presses back. Calls landing in that window
+are **refused** with a result telling the agent to call `read_page` and retry. Most would have
+missed anyway; the guard exists for the case where a same-named control on the new page matches and
+the agent would otherwise act on the wrong page without either side noticing. `read_page` and tools
+marked `x-navigates` are exempt, and the guard is inert when no `getPageMap` is set.
+
 **`registerPageState({ name, read, write?, schema? })`** — ergonomic sugar over `registerTool` for
 SPA app state (Redux/Zustand/signals). It auto-generates a `read_<name>` (read-only) tool and, when
 `write` is supplied, a `set_<name>` tool stamped `x-destructive`: