docs: scaffold agent instruction tree

Add AGENTS.md, CLAUDE.md, and the agents-docs/ tree (workflow, lessons,
engineering standards, context map, ADR seed, feature template) plus a
domain-bearing CONTEXT.md for each of the six subdomains: toju-app,
electron, server, e2e, website, docs-site.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

e2e/CONTEXT.md

@@ -0,0 +1,50 @@
+# End-to-End Suite (e2e)
+
+Owns Playwright-based end-to-end verification of the desktop product. Tests boot the real Electron application against the real signaling server and exercise user-visible flows across chat, voice, screen-share, settings, plugins, and auth.
+
+> **Format reference:**
+> - **Vocabulary** — bold term, one-sentence definition, aliases to avoid.
+> - **Relationships** — bullets with bold terms and cardinality.
+> - **Boundaries / IO** — what this subdomain exposes and consumes.
+> - **Invariants** — rules that always hold.
+> - **Flagged ambiguities** — terms in dispute with proposed resolutions.
+>
+> See `agents-docs/AGENTS_CONTEXT.md` for the contract. Update in the same turn a trigger fires (see `agents-docs/AGENT_WORKFLOW.md` § CONTEXT.md upkeep).
+
+## Vocabulary
+
+| Term | Definition | Aliases to avoid |
+|------|------------|------------------|
+| **Feature area** | A top-level folder under `tests/` (`auth/`, `chat/`, `voice/`, `screen-share/`, `settings/`, `plugins/`) corresponding to a slice of user-visible behavior. | "category", "section" |
+| **Page object** | A test-side abstraction over a screen or panel that exposes user-intent methods rather than raw selectors. | "page model" |
+| **Fixture** | A Playwright `test.extend(...)` setup that prepares one or more user/app instances before a test runs. | "helper" |
+| **Pair test** | An E2E test that boots two Electron instances simultaneously to verify P2P flows (calls, screen-share, transfers). | "multi-client test" |
+
+## Relationships
+
+- A **Feature area** owns one or more `*.spec.ts` files plus its own **Page objects** and **Fixtures**.
+- A **Pair test** depends on the **server** subdomain being reachable — both clients connect to the same signaling server.
+- **Page objects** depend only on the rendered DOM produced by **toju-app**; if a selector changes, only the page object should need updating.
+- The suite as a whole depends on **electron** (built via `npm run build:electron`) and a usable **server** (`npm run server:dev` or `npm run dev`).
+
+## Boundaries / IO
+
+- **Exposes:** test results (HTML report at `test-results/html-report`, JUnit/JSON output on CI). No production surface.
+- **Consumes:**
+  - The built product (toju-app + electron) — typically launched via Playwright's Electron support.
+  - The signaling server (started before the suite runs).
+  - System resources: audio devices for voice tests, screen-capture for screen-share tests. The `.agents/skills/playwright-e2e/SKILL.md` documents how the suite handles the multi-client setup.
+
+## Invariants
+
+- Tests interact only through **Page objects** and **Fixtures** — no raw `page.click('.css-class-name')` scattered across specs.
+- Tests must clean up state between runs — a flaky run that leaves cruft in the local database or signaling server is a bug, not an environment issue.
+- The suite must run headless on CI (`npm run test:e2e`); the `ui` and `debug` variants exist for local development only.
+
+## Flagged ambiguities
+
+- _None recorded yet — add entries when a test concept (e.g. "pair test" vs "multi-client test") resists clean definition._
+
+---
+
+*Agent-procedural rules (TDD, lint) live in `/AGENTS.md`. Practical patterns for writing Playwright tests on this project live in `.agents/skills/playwright-e2e/SKILL.md`. This file is the bounded-context domain artefact for the E2E suite.*