@terreno/mcp 57.20.0 → 58.0.0-beta.1

package/dist/docs/versioned/57.20.0/tasks/support-agent.md

@@ -0,0 +1,230 @@
+# Tasks: Support answering agent (`@terreno/support`)
+
+See: [`docs/implementationPlans/support-agent.md`](../implementationPlans/support-agent.md)
+
+**Status:** Draft (Grow). Do not Pick until the IP header is **Approved**.
+
+## Instructions for the implementing agent
+
+- Load `update-docs`, `terreno-backend-api`, `backend-test-env`, `mongoose-schema-safety`,
+  and `ai-prompt-governance` (prompt tasks). Load `model-router-actions` for every route.
+- TDD: failing Bun test first, then implementation. Mock `LanguageModel` /
+  `EmbeddingModel` the way `ai/src/service/aiService.test.ts` does; never call a real
+  provider in tests.
+- Every task that changes exports or behavior updates the named docs in the same task.
+- Run `bun test --only-failures support/` (and `example-backend/` for tracer tasks), then
+  `bun run lint` on touched packages. `bun run check:knip` before Brew.
+- No `app.get` / `app.post`: routes are `modelRouter` actions. MCP surface is
+  `registerMCPTool` only — no prompts/resources (Draft upstream IP).
+- Luxon for dates; prompt strings are constants at the top of `support/src/agent/prompts.ts`.
+
+---
+
+### Phase 1: Knowledge core
+
+- [ ] **Task 1.1**: Scaffold `@terreno/support` package
+  - Delivers: `support/` workspace package that compiles and runs an empty test; root scripts
+    `support:compile` / `support:test`; package listed in `AGENTS.md` / `CLAUDE.md` / `.cursor/rules/00-root.mdc`
+  - Files: `support/package.json` (deps `@terreno/api`, `@terreno/ai`, `ai`, `zod`, `luxon`, `gray-matter` or equivalent frontmatter parser via catalog), `support/tsconfig.json`, `support/biome.jsonc`, `support/README.md`, `support/src/index.ts`, `support/src/tests/bunSetup.ts`, root `package.json`, `knip.jsonc` (only if a narrow exception is needed), `AGENTS.md`, `CLAUDE.md`, `.cursor/rules/00-root.mdc`
+  - Blocked by: none
+  - Docs: `docs/reference/support.md` (create: header + install snippet), `docs/reference/README.md` link
+  - Skills: `terreno-backend-api`, `update-docs`
+  - Acceptance: `bun run support:compile` and `bun run support:test` succeed; `bun run check:knip` clean; package resolvable from `example-backend` as `workspace:*`
+
+- [ ] **Task 1.2**: Knowledge types + models
+  - Delivers: exported `KnowledgeDocument`, `KnowledgeSource`, `Retriever`, `RetrievedChunk`; Mongoose models `SupportDocument`, `SupportChunk` with indexes from the IP (unique `{sourceId, externalId}`, `{documentId, order}`, text index on chunk `text` + `heading`)
+  - Files: `support/src/knowledge/types.ts`, `support/src/models/supportDocument.ts`, `support/src/models/supportChunk.ts`, `support/src/modelTypes.ts`, `support/src/tests/models.test.ts`
+  - Blocked by: Task 1.1
+  - Docs: `docs/reference/support.md` (Models + Interfaces sections)
+  - Skills: `mongoose-schema-safety`, `terreno-backend-api`, `update-docs`
+  - Acceptance: test inserts two documents with the same `{sourceId, externalId}` and the second fails with a duplicate-key error; `SupportChunk.collection.indexes()` includes a text index
+
+- [ ] **Task 1.3**: Markdown chunker
+  - Delivers: `chunkMarkdown({title, body, maxTokens = 800, overlapTokens = 100})` — heading-aware, keeps code fences intact, returns `[{order, heading, text}]`
+  - Files: `support/src/knowledge/chunker.ts`, `support/src/knowledge/chunker.test.ts`
+  - Blocked by: Task 1.1
+  - Docs: `docs/explanation/support-agent.md` (create: Retrieval section, chunking paragraph)
+  - Skills: `update-docs`
+  - Acceptance: a doc with three H2 sections yields ≥ 3 chunks each carrying its heading; a 5,000-token section splits with overlap and no chunk exceeds `maxTokens`; a fenced code block is never split mid-fence
+
+- [ ] **Task 1.4**: `markdownDocsSource` + `SupportIndexer` (lexical only)
+  - Delivers: `markdownDocsSource({dir, id = "docs", watch})` parsing frontmatter (`title`, `slug`, `audience`, `tags`, `related`); `SupportIndexer.reindex({sourceId?})` that upserts `SupportDocument` by hash, replaces chunks only when the hash changed, marks missing docs `stale`, and returns `{indexed, unchanged, stale}`
+  - Files: `support/src/knowledge/sources/markdownDocsSource.ts`, `support/src/knowledge/indexer.ts`, `support/src/tests/indexer.test.ts`, `support/src/tests/fixtures/kb/*.md`
+  - Blocked by: Task 1.2, Task 1.3
+  - Docs: `docs/how-to/add-support-knowledge.md` (create: `support/kb` convention + frontmatter table), `docs/how-to/README.md` link
+  - Skills: `terreno-backend-api`, `update-docs`
+  - Acceptance: **AC2** — first `reindex()` indexes every fixture; second reports all `unchanged` and chunk count is identical; editing one fixture body re-chunks only that document; deleting a fixture marks it `stale`
+
+- [ ] **Task 1.5**: Lexical retriever + `SupportAgent.search`
+  - Delivers: `lexicalRetriever` over `$text` with `sourceIds` filter, `status: "active"` filter, and post-filter through `source.isVisible` / global `isVisible`; `SupportAgent.search({query, user, limit, sourceIds})` returning excerpts via `excerptBody`-style helper
+  - Files: `support/src/knowledge/retrievers/lexicalRetriever.ts`, `support/src/agent/supportAgent.ts`, `support/src/agent/excerpt.ts`, `support/src/tests/search.test.ts`
+  - Blocked by: Task 1.4
+  - Docs: `docs/explanation/support-agent.md` (Visibility section)
+  - Skills: `terreno-backend-api`, `update-docs`
+  - Acceptance: **AC3** — `search("archive todo")` ranks the archive fixture first; a source whose `isVisible` returns false for the test user contributes zero results; `stale` documents never appear
+
+---
+
+### Phase 2: Answering + REST
+
+- [ ] **Task 2.1**: Prompts + `SupportAgent.ask` (mocked model)
+  - Delivers: `SUPPORT_SYSTEM_PROMPT`, `SUPPORT_ANSWER_SCHEMA` (zod) constants; `ask({question, user, conversationId?, channel})` → retrieve → `aiService.generateJsonObject` → `SupportAnswer`; empty retrieval short-circuits to `shouldEscalate: true` without a model call
+  - Files: `support/src/agent/prompts.ts`, `support/src/agent/supportAgent.ts`, `support/src/tests/ask.test.ts`
+  - Blocked by: Task 1.5
+  - Docs: `docs/explanation/support-agent.md` (Answering + grounding section); prompt governance note in `docs/reference/support.md`
+  - Skills: `ai-prompt-governance`, `terreno-backend-api`, `update-docs`
+  - Acceptance: **AC4** — mock model receives a system prompt containing each retrieved chunk's text and nothing from non-retrieved fixtures; returned `sources` ⊆ retrieved documents; **AC5** (model part) — zero hits → no `doGenerate` call, `shouldEscalate: true`
+
+- [ ] **Task 2.2**: `SupportQuestion` log, feedback, escalation hook, conversation history
+  - Delivers: `SupportQuestion` model; every `ask` writes one row with `ownerId` for an authenticated asker, `channel`, `latencyMs`, `aiRequestId`; `onEscalate` invoked once when `shouldEscalate`; `conversationId` threads only the last 5 turns owned by the current authenticated user into the prompt; anonymous asks are single-turn; `recordFeedback({questionId, user, rating, comment})`
+  - Files: `support/src/models/supportQuestion.ts`, `support/src/agent/supportAgent.ts`, `support/src/tests/questionLog.test.ts`
+  - Blocked by: Task 2.1
+  - Docs: `docs/reference/support.md` (SupportQuestion model, hooks table)
+  - Skills: `mongoose-schema-safety`, `terreno-backend-api`, `update-docs`
+  - Acceptance: **AC5** (log + hook part) — escalated ask persists a `SupportQuestion` and calls a spy `onEscalate` exactly once; second ask by the same user with the same `conversationId` includes the first Q/A in the mock's prompt, while another user's ask with that id has no history; feedback stores `{rating, comment, at}`
+
+- [ ] **Task 2.3**: `SupportApp` plugin + REST routes
+  - Delivers: `SupportApp implements TerrenoPlugin` (constructor per IP; boot-time non-blocking `reindex()`); `modelRouter("/support/questions", SupportQuestion, {collectionActions: {ask, search}, instanceActions: {feedback}, permissions, allowAnonymous})`; public access requires `permissions.ask: [Permissions.IsAny]` with `allowAnonymous: true`; `modelRouter("/support/documents", SupportDocument, {collectionActions: {reindex}})` admin-only; OpenAPI registered
+  - Files: `support/src/supportApp.ts`, `support/src/routes.ts`, `support/src/tests/routes.test.ts`, `support/src/index.ts`
+  - Blocked by: Task 2.2
+  - Docs: `docs/reference/support.md` (REST table, constructor options); `docs/explanation/model-router-actions.md` gets a one-line cross-link example
+  - Skills: `model-router-actions`, `terreno-backend-api`, `backend-test-env`, `update-docs`
+  - Acceptance: **AC6** — supertest: unauthenticated `ask` 401; authenticated 200 with `SupportAnswer` shape; `permissions.ask: [Permissions.IsAny]` with `allowAnonymous: true` → anonymous 200; `permissions.ask: []` → 405; non-owner `feedback` 403; non-admin `reindex` 403; admin `reindex` returns indexer summary
+
+- [ ] **Task 2.4**: Example-backend tracer (kb docs + wiring + SDK)
+  - Delivers: `example-backend/support/kb/` with ≥ 3 docs about todos (create, complete, archive); `SupportApp` registered in `server.ts` with `markdownDocsSource`; `AIService` from existing `example-backend/src/api/ai.ts` provider; example-frontend SDK regenerated
+  - Files: `example-backend/support/kb/*.md`, `example-backend/src/api/support.ts`, `example-backend/src/server.ts`, `example-backend/src/api/support.test.ts`, `example-backend/package.json`, `example-frontend/store/openApiSdk.ts` (generated only)
+  - Blocked by: Task 2.3
+  - Docs: `example-backend/README.md` (Support section); `docs/how-to/add-support-knowledge.md` (worked example)
+  - Skills: `terreno-backend-api`, `generate-sdk`, `update-docs`
+  - Acceptance: example-backend test boots the app, waits for reindex, and `POST /support/questions/search {query: "archive"}` returns the archive doc; `bun run sdk` diff contains `usePostSupportQuestionsAskMutation` (or the generated equivalent) and nothing hand-edited
+
+---
+
+### Phase 3: MCP tools
+
+- [ ] **Task 3.1**: `support_ask` + `support_search` tools
+  - Delivers: `registerSupportMCPTools({agent, permissions})` called by `SupportApp.register`; zod schemas per IP; handlers reuse `SupportAgent` with `channel: "mcp"`; permission denial returns `isError: true`
+  - Files: `support/src/mcpTools.ts`, `support/src/tests/mcpTools.test.ts`, `support/src/supportApp.ts`
+  - Blocked by: Task 2.3
+  - Docs: `docs/how-to/expose-mcp-tools.md` (Support tools subsection); `docs/reference/support.md` (MCP tools table)
+  - Skills: `terreno-backend-api`, `update-docs`
+  - Acceptance: **AC7** (part) — MCP integration test (pattern: `api/src/mcp/integration.test.ts`) as authenticated user: `tools/list` contains `support_ask`, `support_search`; `support_ask` returns text JSON parseable into `SupportAnswer` and a `SupportQuestion` row with `channel: "mcp"`; unauthenticated call is rejected per MCP auth
+
+- [ ] **Task 3.2**: `support_get_document` + `support_list_sources`
+  - Delivers: full markdown + metadata for a visible document; sources summary `{id, label, documentCount, lastIndexedAt}`
+  - Files: `support/src/mcpTools.ts`, `support/src/agent/supportAgent.ts` (`getDocument`, `listSources`), `support/src/tests/mcpTools.test.ts`
+  - Blocked by: Task 3.1
+  - Docs: `docs/reference/support.md` (tools table complete)
+  - Skills: `terreno-backend-api`, `update-docs`
+  - Acceptance: **AC7** (rest) — unknown `documentId` → `isError: true`; hidden document (visibility false) → `isError: true` with no content leak; `support_list_sources` counts match `SupportDocument` `active` rows per source
+
+---
+
+### Phase 4: Sources
+
+- [ ] **Task 4.1**: `modelSource` + example-backend `BugReport` tracer
+  - Delivers: `modelSource({id, model, filter, toDocument, isVisible})`; example-backend `BugReport` model (`title`, `description`, `resolution`, `status`) with seed data and `modelSource` over resolved reports
+  - Files: `support/src/knowledge/sources/modelSource.ts`, `support/src/tests/modelSource.test.ts`, `example-backend/src/models/bugReport.ts`, `example-backend/src/modelInterfaces.ts`, `example-backend/src/api/support.ts`, `example-backend/src/scripts/seed-test-data.ts`, `example-backend/src/api/support.test.ts`
+  - Blocked by: Task 1.5
+  - Docs: `docs/how-to/add-support-knowledge.md` (Model source section: bug reports and resolutions)
+  - Skills: `mongoose-schema-safety`, `terreno-backend-api`, `update-docs`
+  - Acceptance: **AC9** — resolved reports indexed, open excluded; `search("crash on login")` returns the resolution excerpt
+
+- [ ] **Task 4.2**: `announcementsSource`
+  - Delivers: adapter over the `Announcement` model: `published` for everyone, `archived` with `includeArchived`, drafts never; `post("save")` hook triggers `reindex({sourceId})` for that document; `isVisible` delegates to an optional `matchAudience` passed in
+  - Files: `support/src/knowledge/sources/announcementsSource.ts`, `support/src/tests/announcementsSource.test.ts` (dev-dependency on `@terreno/announcements` for the test model only), `example-backend/src/api/support.ts`
+  - Blocked by: Task 1.5, **merge of [PR #1284](https://github.com/TerrenoLabs/terreno/pull/1284)**
+  - Docs: `docs/how-to/add-support-knowledge.md` (Announcements section); `docs/reference/announcements.md` (link to support integration)
+  - Skills: `terreno-backend-api`, `update-docs`
+  - Acceptance: **AC8** — status matrix test; publishing an announcement results in a new `SupportDocument` after the save hook resolves, without a manual `reindex()`
+
+- [ ] **Task 4.3**: Reindex lifecycle hardening
+  - Delivers: `watch: true` on `markdownDocsSource` re-indexes a changed file (debounced); boot reindex never blocks `TerrenoApp.start()`; indexer errors from one source do not abort others and are logged with `logger.warn`
+  - Files: `support/src/knowledge/indexer.ts`, `support/src/knowledge/sources/markdownDocsSource.ts`, `support/src/tests/indexerLifecycle.test.ts`
+  - Blocked by: Task 2.3, Task 4.1
+  - Docs: `docs/explanation/support-agent.md` (Indexing lifecycle section)
+  - Skills: `terreno-backend-api`, `update-docs`
+  - Acceptance: a source whose `list()` throws leaves other sources indexed and the summary reports it; file change under `watch` produces a new `contentHash` within the debounce window in a test using a temp dir
+
+---
+
+### Phase 5: Embeddings + hybrid retrieval
+
+- [ ] **Task 5.1**: `embeddingModel` option + chunk embeddings
+  - Delivers: indexer calls `embedMany` in batches when `embeddingModel` is set; stores `embedding` + `embeddingModelId`; re-embeds only when the model id differs or the chunk is new; embedding failure logs `warn` and leaves chunks lexical-only
+  - Files: `support/src/knowledge/indexer.ts`, `support/src/knowledge/embeddings.ts`, `support/src/tests/embeddings.test.ts`
+  - Blocked by: Task 1.4
+  - Docs: `docs/reference/support.md` (options: `embeddingModel`); `docs/reference/environment-variables.md` (example-backend key used for embeddings)
+  - Skills: `terreno-backend-api`, `update-docs`
+  - Acceptance: **AC10** (storage part) — mocked `EmbeddingModel` yields embeddings on every chunk; switching `modelId` re-embeds all; no `embeddingModel` → no `embedding` field
+
+- [ ] **Task 5.2**: Vector + hybrid retrievers with RRF
+  - Delivers: `vectorRetriever` (in-process cosine; `$vectorSearch` when `atlasVectorSearchIndex` set), `hybridRetriever` fusing lexical + vector by reciprocal rank fusion; `SupportApp` picks hybrid when embeddings exist, lexical otherwise; `retriever` option overrides
+  - Files: `support/src/knowledge/retrievers/vectorRetriever.ts`, `support/src/knowledge/retrievers/hybridRetriever.ts`, `support/src/supportApp.ts`, `support/src/tests/hybridRetriever.test.ts`
+  - Blocked by: Task 5.1, Task 1.5
+  - Docs: `docs/explanation/support-agent.md` (Retrieval: lexical vs hybrid vs Atlas); `docs/reference/support.md` (`atlasVectorSearchIndex`, `retriever`)
+  - Skills: `terreno-backend-api`, `update-docs`
+  - Acceptance: **AC10** (ranking part) — with deterministic mock embeddings a chunk with zero lexical overlap but nearest vector lands in top 3; custom `retriever` is called instead of the default; `$vectorSearch` path is unit-tested by asserting the aggregation pipeline shape (no Atlas in tests)
+
+- [ ] **Task 5.3**: Example-backend embedding wiring
+  - Delivers: `example-backend` passes `google.textEmbedding(...)` when the Google key is present (reuse provider loading in `example-backend/src/api/ai.ts`), otherwise omits `embeddingModel`; README documents both modes
+  - Files: `example-backend/src/api/support.ts`, `example-backend/src/api/ai.ts`, `example-backend/README.md`
+  - Blocked by: Task 5.2, Task 2.4
+  - Docs: `docs/reference/environment-variables.md`; `example-backend/README.md`
+  - Skills: `terreno-backend-api`, `update-docs`
+  - Acceptance: example-backend test with env unset boots lexical-only and `search` still works; a unit test asserts the embedding model is constructed only when the key is set
+
+---
+
+### Phase 6: AI authoring loop
+
+- [ ] **Task 6.1**: `terreno-support check` CLI
+  - Delivers: bin `terreno-support` with `check [dir]` validating frontmatter (`title`, `slug`), duplicate slugs, empty bodies, zero-chunk files; JSON and human output; non-zero exit on failure
+  - Files: `support/src/cli.ts`, `support/src/knowledge/validateKb.ts`, `support/src/tests/cli.test.ts`, `support/package.json` (`bin`), example-backend `package.json` script `support:check`
+  - Blocked by: Task 1.4
+  - Docs: `docs/how-to/add-support-knowledge.md` (Validate section); `docs/reference/support.md` (CLI)
+  - Skills: `update-docs`
+  - Acceptance: **AC11** — fixtures for pass and each failure class; failing output names the file and rule
+
+- [ ] **Task 6.2**: `write-support-docs` skill
+  - Delivers: plugin skill instructing agents to add/update `support/kb/<slug>.md` for user-visible changes, run `terreno-support check`, mention the doc in the PR; Roast criterion text; generated Claude copy via `bun run skills:sync`
+  - Files: `plugins/terreno-planning/skills/write-support-docs/SKILL.md`, generated `plugins/terreno-claude/skills/write-support-docs/SKILL.md`, `skills/` output, `plugins/README.md` skill list, `.rulesync/rules` or Pick guidance reference (`plugins/terreno-planning/skills/terreno-2-pick/SKILL.md` supporting-skills mention)
+  - Blocked by: Task 6.1
+  - Docs: `docs/reference/lifecycle-plugin.md` (skill table row); `docs/how-to/install-agent-skills.md` if it enumerates skills
+  - Skills: `update-agent-docs`, `update-docs`
+  - Acceptance: **AC14** — `bun run skills:sync` produces the generated copies with no manual edits; `bun run check:lifecycle-skills` passes; skill has a reliable trigger line and completion criteria per `update-agent-docs`
+
+- [ ] **Task 6.3**: Knowledge gaps + admin contribution
+  - Delivers: `GET /support/questions/gaps` collection action grouping `low` confidence and `down` feedback by normalized question (`{normalizedQuestion, count, lastAskedAt, sampleQuestions, sampleQuestionIds}`); `adminContribution()` registering `SupportQuestion` and `SupportDocument` read-only
+  - Files: `support/src/routes.ts`, `support/src/agent/gaps.ts`, `support/src/supportApp.ts`, `support/src/tests/gaps.test.ts`
+  - Blocked by: Task 2.3
+  - Docs: `docs/how-to/add-support-knowledge.md` (Close the loop: gaps → new doc); `docs/reference/support.md` (gaps action, admin models)
+  - Skills: `model-router-actions`, `building-admin-interfaces`, `update-docs`
+  - Acceptance: **AC12** — two `low` asks with case/punctuation variants collapse into one row with `count: 2`; non-admin 403; admin app lists both models
+
+---
+
+### Phase 7: Docs consolidation
+
+- [ ] **Task 7.1**: Explanation, reference, how-to reconciliation + Notion how-to
+  - Delivers: complete `docs/explanation/support-agent.md` (why RAG in-process, source contract, visibility, authoring loop, relation to announcements help API and hosted MCP), complete `docs/reference/support.md`, `docs/how-to/plug-in-notion-knowledge.md` with a full `KnowledgeSource` implementation using `@notionhq/client` that typechecks in a docs compile test, index links in all three READMEs, root `README.md` package table
+  - Files: the four docs pages, `docs/{explanation,reference,how-to}/README.md`, `README.md`, `support/src/tests/docsSnippets.test.ts` (compiles the Notion snippet against exported types with a stubbed client)
+  - Blocked by: Task 3.2, Task 4.2, Task 5.2, Task 6.3
+  - Docs: all of the above
+  - Skills: `update-docs`, `docs-audit`
+  - Acceptance: **AC13** — pages exist and are linked; `docs-audit` reports no drift for `@terreno/support`; Notion snippet compiles; no page documents prompts/resources as shipped
+
+---
+
+## Frontier after Approve
+
+Unblocked: **Task 1.1**.  
+After 1.1: 1.2, 1.3 in parallel.  
+1.4 after 1.2 + 1.3; 1.5 after 1.4.  
+Phase 2 is linear (2.1 → 2.2 → 2.3 → 2.4).  
+3.1 after 2.3; 3.2 after 3.1.  
+4.1 and 5.1 can start after 1.5 / 1.4 in parallel with Phase 2.  
+4.2 additionally waits for PR #1284 to merge — if it has not merged when the frontier reaches it, skip it and continue; Brew notes the gap.  
+6.1 after 1.4; 6.2 after 6.1; 6.3 after 2.3.  
+7.1 last.

package/dist/docs/versioned/57.20.0/tasks/ui-performance.md

@@ -0,0 +1,134 @@
+# Task List: UI Render and Startup Performance
+
+**Feature profile:** true
+
+## Durable contract
+
+- **Destination:** `@terreno/ui` avoids unnecessary render, parse, and layout work on common app
+  paths without changing visible output, interactions, accessibility, or public root imports.
+- **Completed foundation:** P0 primitives and theme work landed in PR #1131. P1 table-update,
+  icon-update, and package-subpath work is covered by `bun run performance:p1` and
+  `docs/reference/ui-performance.md`.
+- **Scope:** the remaining static-audit priorities below, in priority order. Each optimization must
+  add a representative example renderer or import workload, capture a same-machine baseline, and
+  add behavior regression coverage before production code changes.
+- **Non-goals:** visual redesigns, public API removals, root-barrel removal, or benchmark-driven
+  changes without a measured improvement.
+- **Decisions:** keep `@terreno/ui` root imports supported; use Luxon for date work; preserve
+  React Native Web; treat changed-prop/theme work as required behavior rather than overhead to hide.
+- **Tracer:** one real demo story and one benchmark workload per task, exercised before and after
+  the optimization.
+- **Verification seam:** focused Bun regressions, `bun run ui:test`, `bun run lint`,
+  `bun run compile`, the relevant performance script, and a running demo video.
+
+## Instructions for the implementing agent
+
+- Work one task per PR. Do not combine unrelated hot paths.
+- Add regression coverage and record the baseline before changing production code.
+- Report initial, equivalent-prop, and changed-prop medians. Startup tasks must use fresh-process or
+  bundle-graph measurements so module caches do not invalidate the comparison.
+- A faster result with changed output, lost keyboard behavior, stale theme data, or stale layout is
+  a failure.
+- Update `docs/reference/ui-performance.md` with the final same-machine measurements.
+
+### Phase 1: Finish omitted and deferred high-priority work
+
+- [x] **Task 1.1**: Stabilize `OpenAPIProvider` context values (leftover P0)
+  - Delivers: unrelated parent renders no longer fan out through unchanged OpenAPI model metadata.
+  - Files: `ui/src/OpenAPIContext.tsx`, focused regression tests, `ui/benchmarks/`, `ui/package.json`,
+    `docs/reference/ui-performance.md`
+  - Blocked by: none
+  - Acceptance: benchmark an example tree with many OpenAPI consumers; memoize provider actions and
+    value with complete dependencies; unchanged metadata skips consumer renders; replacing the spec
+    updates field lookups immediately; existing form/admin behavior and the full UI suite pass.
+
+- [x] **Task 1.2**: Virtualize large `DataTable` row sets without changing table behavior
+  - Delivers: initial mount cost and mounted row count scale with the viewport instead of total rows.
+  - Files: `ui/src/DataTable.tsx`, `ui/src/DataTable.test.tsx`, P1 regressions and benchmark,
+    `demo/stories/DataTable.stories.tsx`, `docs/reference/ui-performance.md`
+  - Blocked by: none
+  - Acceptance: benchmark at least 1,000 rows with pinned and unpinned columns; mounted rows remain
+    viewport-bounded; sorting, pagination, alternating/highlight colors, custom cells, row test IDs,
+    header/body scroll sync, pinned columns, and row-detail modals match current behavior on web and
+    native; changed rows render correctly after scrolling.
+
+- [ ] **Task 1.3**: Keep optional `DataTable` markdown off the default import path
+  - Delivers: tables without `infoModalText` do not parse or evaluate the markdown renderer.
+  - Files: `ui/src/DataTable.tsx`, a focused header-info component, tests, import benchmark,
+    `docs/reference/ui-performance.md`
+  - Blocked by: Task 1.2
+  - Acceptance: a fresh-process or bundle-graph measurement proves markdown is absent from the
+    default table path; opening header info still renders identical markdown with an explicit
+    loading/error state; table snapshots and interactions remain unchanged.
+
+- [x] **Task 1.4**: Measure and isolate remaining heavy root-import modules
+  - Delivers: startup guidance and lazy boundaries for heavy optional widgets while the root import
+    remains compatible.
+  - Files: `ui/src/index.tsx`, heavy components identified by the measurement, `ui/package.json`,
+    import benchmark, `docs/reference/ui.md`, `docs/reference/ui-performance.md`
+  - Blocked by: none
+  - Acceptance: compare fresh root and subpath imports by evaluated module count, output bytes, and
+    elapsed time; isolate only measured heavy optional paths; `import {Button} from "@terreno/ui"`
+    remains valid; every existing `@terreno/ui/*`, `src/*`, and `dist/*` path still resolves.
+
+- [ ] **Task 1.5**: Decide whether table chrome needs a smaller icon path
+  - Delivers: an evidence-based keep-or-change decision for FontAwesome table chrome.
+  - Files: `ui/src/Icon.tsx`, `ui/src/DataTable.tsx`, icon/import benchmarks, tests,
+    `docs/reference/ui-performance.md`
+  - Blocked by: Task 1.3
+  - Acceptance: measure current glyph import and render cost first; if a smaller path materially
+    improves it, preserve icon names, sizes, weights, colors, custom-icon precedence, font loading,
+    and accessibility; otherwise make no production change and record the measured no-change
+    decision.
+
+### Phase 2: P2 component hot paths
+
+- [ ] **Task 2.1**: Stabilize `DateTimeField` sheet identity and cold platform paths
+  - Delivers: rerenders retain the same sheet ref, and unused desktop/mobile date-picker paths stay
+    cold.
+  - Files: `ui/src/DateTimeField.tsx`, date-time sheet modules, tests, benchmark/example renderer,
+    `docs/reference/ui-performance.md`
+  - Blocked by: Task 1.4
+  - Acceptance: replace render-time `createRef` with a stable ref; benchmark closed and open states;
+    date, time, datetime, timezone, validation, open/close, and imperative behavior match on web and
+    native; first-open loading and import failures have tested states.
+
+- [ ] **Task 2.2**: Load emoji data only when `EmojiSelector` opens
+  - Delivers: apps that do not open the selector avoid parsing `emoji-datasource`.
+  - Files: `ui/src/EmojiSelector.tsx`, tests, import benchmark/example renderer,
+    `docs/reference/ui-performance.md`
+  - Blocked by: Task 1.4
+  - Acceptance: fresh-process or bundle-graph results prove emoji data is absent before open;
+    first open has tested loading/error states; search, categories, selection, skin-tone behavior,
+    reopen behavior, and web/native rendering remain unchanged.
+
+- [x] **Task 2.3**: Stabilize plain and confirmation `Button` render paths
+  - Delivers: debounce identity and plain-button renders stay stable without weakening duplicate
+    submission protection.
+  - Files: `ui/src/Button.tsx`, button tests, benchmark/example renderer,
+    `docs/reference/ui-performance.md`
+  - Blocked by: none
+  - Acceptance: equivalent props skip avoidable work; plain buttons do not allocate confirmation
+    state or load the modal path; rapid presses, delayed handlers, unmount cleanup, loading,
+    disabled, keyboard, haptic, and confirmation behavior match current tests and running stories.
+
+- [x] **Task 2.4**: Replace synchronous responsive dimension reads with reactive breakpoints
+  - Delivers: responsive `Box` props read one shared window state and update correctly on rotation
+    and resize.
+  - Files: `ui/src/MediaQuery.ts`, `ui/src/Box.tsx`, provider/hook files if needed, tests,
+    benchmark/example renderer, `docs/reference/ui-performance.md`
+  - Blocked by: none
+  - Acceptance: benchmark a large responsive `Box` tree; remove per-box synchronous
+    `Dimensions.get` calls; native `sm`/`md`/`lg`/`xl` (320/375/600/1024) and web `lg`/`xl`
+    (1024/1280) boundaries match the product spec; resize, rotation, SSR hydration, web, and
+    native behavior have regressions and running-demo proof.
+
+- [ ] **Task 2.5**: Close the measured hot-style allocation backlog
+  - Delivers: only proven hot leaves reuse style objects without stale theme values.
+  - Files: measured hotspots in `DataTable`, `Button`, `Modal`, `Filter`, `Text`, or `Box`; focused
+    tests and benchmarks; `docs/reference/ui-performance.md`
+  - Blocked by: Tasks 1.2, 2.3, 2.4
+  - Acceptance: profile first and list the selected hotspots; cache styles with every theme/layout
+    dependency represented; equivalent renders improve measurably; changed themes, dimensions,
+    highlights, disabled states, and animations remain visually identical; do not perform a
+    library-wide style rewrite.

package/dist/docs/versioned/57.20.0/tasks/unified-mutation-executors.md

@@ -0,0 +1,35 @@
+# Tasks: Unified mutation executors
+
+IP: [unified-mutation-executors.md](../implementationPlans/unified-mutation-executors.md)
+
+## Phase 1 — MCP create → `executeCreate`
+
+- [x] **Task 1.1**: MCP create uses `executeCreate`
+  - Delivers: `handleCreate` calls `executeCreate`; `model.create` gone from create path; MCP error mapping for `APIError`
+  - Files: `api/src/mcp/handlers.ts`, `api/src/mcp/integration.test.ts`, `api/src/sync/executors.test.ts` (only if a shared hook case is missing)
+  - Blocked by: none
+  - Docs: none yet (behavior still MCP-only on create)
+  - Skills: `backend-test-env` if tests touch `process.env`; `update-docs` not required until Phase 3
+  - Acceptance: existing MCP `handleCreate` tests pass; one test proves `preCreate` runs for MCP create; `rg "model.create" api/src/mcp/handlers.ts` is empty on the create path
+
+## Phase 2 — Update and delete
+
+- [x] **Task 2.1**: MCP update uses `executeUpdate`
+  - Delivers: `handleUpdate` adapter only
+  - Files: `api/src/mcp/handlers.ts`, `api/src/mcp/integration.test.ts`
+  - Blocked by: 1.1
+  - Acceptance: MCP update tests pass; no `doc.save()` in `handleUpdate`
+- [x] **Task 2.2**: MCP delete uses `executeDelete`
+  - Delivers: `handleDelete` adapter only (soft-delete semantics from executor)
+  - Files: `api/src/mcp/handlers.ts`, `api/src/mcp/integration.test.ts`
+  - Blocked by: 1.1
+  - Acceptance: MCP delete tests pass; persistence only inside executor
+
+## Phase 3 — Contract and docs
+
+- [x] **Task 3.1**: Docs + changelog
+  - Delivers: explanation/reference match the single write pipeline
+  - Files: `docs/explanation/model-router-actions.md`, `docs/reference/api.md`, `docs/implementationPlans/model-router-mcp.md` (pointer), changelog fragment
+  - Blocked by: 2.1, 2.2
+  - Skills: `update-docs`
+  - Acceptance: a stranger can find “MCP writes use executors” from docs without reading the PR

package/dist/docs/versioned/57.20.0/tasks/upgrade-guides-and-skill.md

@@ -0,0 +1,118 @@
+# Task List: Upgrade Guides and the `upgrading-terreno` Skill
+
+See: [`docs/implementationPlans/upgrade-guides-and-skill.md`](../implementationPlans/upgrade-guides-and-skill.md)
+
+**RTK deprecation flag:** **Blocked.** The RTK → syncdb upgrade is the primary use case for this skill. Do not start until PR #869 merges and [`rtk-to-syncdb-migration-docs`](rtk-to-syncdb-migration-docs.md) Phase 3 has produced the migration guide the skill will delegate to. Tasks marked `[RTK]` specifically handle that migration.
+
+## Instructions for the implementing agent
+
+- Derive upgrade-note content from evidence: `gh release view <tag>`, `git log <prev>..<tag>`, and diffs of public API surfaces (`*/src/index.ts`). Do not write "possibly breaking" — determine whether it breaks by reading the diff.
+- Read `.rulesync/skills/upgrading-expo/SKILL.md` before writing the new skill. Do not duplicate its content; invoke it.
+- The skill must never leave a consumer's repo in a half-upgraded state without saying so. Every failure path must report what succeeded, what failed, and how to roll back.
+- Run `bun run mcp:build` after changing anything under `mcp-server/`. Run `bun run rules:check` after touching `.rulesync/`.
+
+## Phase 1: Note format and backfill
+
+- [x] **Task 1.1**: Define the upgrade note format
+  - Description: Create `mcp-server/src/docs/upgrades/README.md` documenting the required note format from the IP: frontmatter-style header (`Action required`, `Affected packages`) then `Summary`, `Breaking changes` (each with What changed / Why / Migration with before-and-after code), `Deprecations`, `New capabilities`, `Verification`. Explain that notes are concatenated by `terreno_get_upgrade_guide` across a version range, so each must be self-contained and must not assume the reader read an adjacent note. Include a copy-paste template. Read the existing `0.20.0.md` and `0.21.0.md` first and keep the format compatible with them.
+  - Files: `mcp-server/src/docs/upgrades/README.md` (new)
+  - Depends on: none
+  - Acceptance: the format is compatible with the two existing notes; the template includes every required section; the self-contained requirement is stated with its reason.
+
+- [x] **Task 1.2**: Research what changed between 0.21.0 and the current version
+  - Description: For each release from 0.22.0 through the current version, run `gh release view <tag>` and `git log --oneline <prev>..<tag>`, and diff the public surface of each published package (`git diff <prev>..<tag> -- '*/src/index.ts'`). Identify actual breaking changes, deprecations, and notable additions. Record findings per version in the PR body. Pay particular attention to the OpenFeature feature-flag migration, which the README documents as requiring consumer action — that is a breaking change with no upgrade note.
+  - Files: none (findings in the PR body)
+  - Depends on: Task 1.1
+  - Acceptance: every version from 0.22.0 to current is covered; each finding cites a commit or diff; the feature-flag migration is captured.
+
+- [x] **Task 1.3**: Write the consolidated backfill note
+  - Description: Per IP question U1, write a single consolidated note covering 0.21.0 → current (name it for the current version). Include everything from Task 1.2 that requires consumer action, organized by concern rather than by version. Open with an honest note that per-release notes were not maintained for this range and that this note consolidates them. Move the OpenFeature feature-flag migration content out of `README.md` into this note (or into `docs/how-to/add-feature-flags.md`) and link it, rather than leaving a migration guide in the README.
+  - Files: `mcp-server/src/docs/upgrades/<current-version>.md` (new), `README.md`, `docs/how-to/add-feature-flags.md`
+  - Depends on: Task 1.2
+  - Acceptance: every action-requiring change from Task 1.2 appears; the feature-flag migration is no longer in `README.md`; `terreno_get_upgrade_guide` with the range 0.21.0 → current returns this note's content.
+
+- [x] **Task 1.4**: Improve the empty-range response
+  - Description: Update `terreno_get_upgrade_guide` in `mcp-server/src/tools.ts` so that a range with no recorded notes returns an explicit message naming the versions with no notes, rather than empty output — an agent seeing empty output concludes nothing changed, which is the opposite of the truth. Also have it list which versions in the range *did* have notes. Add tests for: range fully covered, range partially covered, range with no notes, and an invalid range.
+  - Files: `mcp-server/src/tools.ts`, tests under `mcp-server/src/`
+  - Depends on: Task 1.3
+  - Acceptance: all four test cases pass; the no-notes response names the specific versions; `bun run mcp:build` succeeds.
+
+## Phase 2: Versioning policy
+
+- [x] **Task 2.1**: Write `docs/explanation/versioning-policy.md`
+  - Description: Document: what lockstep versioning means (all `@terreno/*` packages share a version; you upgrade them together and mixing versions is unsupported — verify this claim against `publish-on-tag.yml` and any peer-dependency constraints before asserting it); the pre-1.0 policy (breaking changes may land in minor releases, always with an upgrade note); the deprecation window from IP question U6 (minimum three minor releases, with the RTK deprecation as the worked example); where to find upgrade notes and the changelog; and what 1.0 will mean when it happens (link program question P9). Keep it short and concrete — this document is a promise, so every sentence must be one the team will keep.
+  - Files: `docs/explanation/versioning-policy.md` (new), `docs/explanation/README.md`
+  - Depends on: Task 1.3
+  - Acceptance: the lockstep claim is verified against the publish workflow and peer deps; the deprecation window is a specific number; the RTK deprecation is used as the example; linked from the explanation index.
+
+- [x] **Task 2.2**: `[RTK]` Write `docs/how-to/upgrade-terreno.md`
+  - Description: The human-facing upgrade guide. Cover: checking your current versions; finding the notes for your range (both via `terreno_get_upgrade_guide` and by browsing the notes directory); the correct ordering with the reasoning for each step (backend before frontend because the OpenAPI spec is the contract; Expo before Terreno frontend packages because `@terreno/ui` pins peers against Expo versions — verify both claims); the manual steps for each phase; verification; and rollback. Add a prominent section for the RTK → syncdb upgrade linking `docs/how-to/migrate-rtk-to-syncdb.md` as a larger, separate operation that should not be combined with a routine version bump.
+  - Files: `docs/how-to/upgrade-terreno.md` (new), `docs/how-to/README.md`
+  - Depends on: Task 2.1
+  - Acceptance: both ordering claims verified against source or `package.json` peer deps and cited; the RTK migration is explicitly separated from routine upgrades; rollback steps present.
+
+## Phase 3: The skill
+
+- [x] **Task 3.1**: Author the `upgrading-terreno` skill
+  - Description: Create `.rulesync/skills/upgrading-terreno/SKILL.md`. Frontmatter: `name: upgrading-terreno`, a `description` naming trigger phrases ("upgrade Terreno", "bump @terreno packages", "update to the latest Terreno", "upgrade terreno version"), `targets: ['*']`. Body: Preconditions (clean git tree — refuse to start otherwise; on a branch, not the default branch; tests currently passing so failures are attributable); Determine versions (`application_info` from the local MCP server, or read `package.json` files); Fetch notes (`terreno_get_upgrade_guide`); **Plan and confirm** — print current version, target version, the notes found, the packages affected, and the ordered steps, then stop for confirmation; then the ten-step ordering from the IP with a verification gate between each phase; Expo delegation (invoke the `upgrading-expo` skill at step 6 rather than duplicating it); Failure handling (report what succeeded, what failed, and the rollback command; on a multi-version jump failure, retry version by version to isolate); and a final report. Include an explicit prohibition on continuing past a failed compile or test run.
+  - Files: `.rulesync/skills/upgrading-terreno/SKILL.md` (new)
+  - Depends on: Task 2.2
+  - Acceptance: the clean-tree precondition is a hard refusal, not a warning; the confirmation gate precedes the first mutation; step 6 invokes `upgrading-expo` by name; the failure path names a rollback command; the ordering matches the how-to guide exactly.
+
+- [x] **Task 3.2**: Add the ordering reference
+  - Description: Create `.rulesync/skills/upgrading-terreno/references/ordering.md` containing the dependency reasoning in detail: why backend precedes frontend, why the typed client is regenerated only after the backend is upgraded and running, why Expo precedes `@terreno/ui`, which packages are backend versus frontend (enumerate them from `publish-on-tag.yml`), and what breaks if the order is violated — with the specific symptom for each violation (for example: regenerating the client against an old backend produces a client missing the new routes, which fails at compile time in the consumer's screens).
+  - Files: `.rulesync/skills/upgrading-terreno/references/ordering.md` (new)
+  - Depends on: Task 3.1
+  - Acceptance: every package is classified backend or frontend and the list matches `publish-on-tag.yml`; each ordering rule has a concrete violation symptom.
+
+- [x] **Task 3.3**: `[RTK]` Add the syncdb migration branch
+  - Description: Add a section to the skill for the case where the upgrade range crosses the syncdb release. The skill must detect this (the range includes the syncdb version), stop, and tell the user that this is a platform migration rather than a version bump, pointing at `docs/how-to/migrate-rtk-to-syncdb.md` and offering to perform the routine version bump first and the migration as a separate operation. Do not attempt to automate the full migration inside a version-bump flow.
+  - Files: `.rulesync/skills/upgrading-terreno/SKILL.md`
+  - Depends on: Task 3.1, `rtk-to-syncdb-migration-docs` Phase 3
+  - Acceptance: the skill detects a range crossing the syncdb version and stops with the two-option choice; it does not attempt the data-layer migration automatically.
+
+- [x] **Task 3.4**: Wire the MCP prompt to the skill
+  - Description: Update the `terreno_upgrade` prompt in `mcp-server/src/prompts.ts` so it points at the `upgrading-terreno` skill as the executor, keeping the prompt as the discovery surface for agents without skill support. Ensure the prompt's ordering matches the skill's exactly so the two cannot drift into disagreement — ideally the prompt references the ordering rather than restating it.
+  - Files: `mcp-server/src/prompts.ts`
+  - Depends on: Task 3.1
+  - Acceptance: the prompt names the skill; the two do not state conflicting orderings; `bun run mcp:build` succeeds.
+
+- [x] **Task 3.5**: Generate skill mirrors
+  - Description: Run `bun run rules` and commit all generated files.
+  - Files: generated skill mirrors
+  - Depends on: Task 3.1, Task 3.2, Task 3.3
+  - Acceptance: `bun run rules:check` exits 0; `upgrading-terreno` appears under every configured target directory.
+
+## Phase 4: Enforcement
+
+- [x] **Task 4.1**: Require upgrade notes in the release skill
+  - Description: Add a required step to `.rulesync/skills/release/SKILL.md`: after assembling the changelog section for the release, if it contains a `Breaking`, `Deprecated`, `Removed`, or `Changed` entry, write `mcp-server/src/docs/upgrades/<version>.md` using the template from Task 1.1 before tagging. Reference the note from the changelog section. Regenerate mirrors.
+  - Files: `.rulesync/skills/release/SKILL.md`, generated mirrors
+  - Depends on: Task 1.1
+  - Acceptance: the step lists the four triggering changelog headings; `bun run rules:check` exits 0.
+
+- [x] **Task 4.2**: Add the CI enforcement check
+  - Description: Add a check (script plus a job in `.github/workflows/repo-policies.yml`, or a step in `publish-on-tag.yml` that runs before publishing) that reads the changelog section for the tag being released and fails if it contains a breaking/deprecated/removed/changed entry while `mcp-server/src/docs/upgrades/<version>.md` does not exist. Write it as a Bun TypeScript script with `const` arrow functions and explicit return types, following the repo's script conventions. Include tests for: note present and required, note missing and required, note not required.
+  - Files: `scripts/check-upgrade-notes.ts` (new), `package.json`, `.github/workflows/repo-policies.yml` or `.github/workflows/publish-on-tag.yml`, tests
+  - Depends on: Task 4.1
+  - Acceptance: all three test cases pass; the check fails a simulated release with a breaking changelog entry and no note; `bun run lint` passes.
+
+- [x] **Task 4.3**: Link notes from the changelog
+  - Description: Update `CHANGELOG.md` so each version section with an upgrade note links it. Add a line to the changelog header explaining that upgrade notes live in `mcp-server/src/docs/upgrades/` and are also available through the MCP server's `terreno_get_upgrade_guide` tool.
+  - Files: `CHANGELOG.md`
+  - Depends on: Task 1.3
+  - Acceptance: every version with a note links it; the header explains where notes live and how to fetch them.
+
+## Phase 5: Validation
+
+- [x] **Task 5.1**: Perform a real multi-version upgrade with the skill
+  - Description: Create a consumer app pinned to an older Terreno version (at least two minors behind current — scaffold with `terreno_bootstrap_app` from that version, or check out `example-frontend`/`example-backend` at the older tag into a scratch directory). Run the `upgrading-terreno` skill to bring it to current. Verify: the clean-tree precondition fires when the tree is dirty, the confirmation gate fires, each phase's verification runs, and the app compiles and runs afterward. Record every gap and fix the skill, the notes, or the how-to as appropriate. Capture the transcript and the running upgraded app to `/opt/cursor/artifacts/`.
+  - Files: `.rulesync/skills/upgrading-terreno/SKILL.md`, `mcp-server/src/docs/upgrades/*`, `docs/how-to/upgrade-terreno.md`, generated mirrors
+  - Depends on: Task 3.5, Task 1.3
+  - Acceptance: the upgrade succeeds and the app runs; both gates observed firing; every gap fixed; artifacts captured; `bun run rules:check` exits 0.
+
+- [x] **Task 5.2**: Verify the failure-isolation path
+  - Description: Deliberately create a failing upgrade (for example by pinning a consumer file to an API removed in the range) and confirm the skill stops at the failing phase, names it, reports what succeeded, offers rollback, and — for a multi-version range — retries version by version to isolate the breaking version. Fix the skill if any of those behaviors is missing.
+  - Files: `.rulesync/skills/upgrading-terreno/SKILL.md`, generated mirrors
+  - Depends on: Task 5.1
+  - Acceptance: all four failure behaviors observed in the transcript; the isolation retry correctly identifies the breaking version; `bun run rules:check` exits 0.

package/dist/docs/versioned/57.20.0/tutorials/README.md

@@ -0,0 +1,11 @@
+# Tutorials
+
+Learning-oriented, hands-on lessons. Start here if you are new to Terreno.
+
+## Contents
+
+- [Getting started](getting-started.md) — Scaffold a new app with `create-terreno-app`, run it locally, or explore monorepo examples
+
+## Next steps
+
+- [Set up Terreno MCP](../how-to/set-up-terreno-mcp.md) — Connect your editor to the hosted `terreno` server and the local `terreno-mcp-local` server (logs, last error, client state, database)

package/dist/docs/versioned/57.20.0/tutorials/getting-started.md

@@ -0,0 +1,84 @@
+# Getting Started
+
+Create a new Terreno app with the scaffolding CLI, then run it locally.
+
+## Prerequisites
+
+- [Bun](https://bun.sh/) installed
+- MongoDB as a **replica set** (required for sync and realtime — see [deployment baseline](../explanation/deployment-baseline.md))
+
+## 1. Scaffold a new app
+
+From an empty parent directory:
+
+```bash
+bunx create-terreno-app my-app --display-name "My App"
+```
+
+npm alternative:
+
+```bash
+npm create terreno-app my-app -- --display-name "My App"
+```
+
+This writes `my-app/backend` and `my-app/frontend` plus deploy and agent config. The CLI does not install dependencies or seed data.
+
+Full flag reference, layout, and deploy notes: [Create a Terreno app](../how-to/create-a-terreno-app.md).
+
+## 2. Install dependencies
+
+```bash
+cd my-app
+(cd backend && bun install)
+(cd frontend && bun install)
+```
+
+## 3. Start MongoDB
+
+Use Atlas or a local single-node replica set. The scaffolded `backend/.env` points at `mongodb://127.0.0.1:27017/my_app?replicaSet=rs0` (database name derived from `appName`).
+Follow [Start MongoDB (replica set)](../how-to/create-a-terreno-app.md#3-start-mongodb-replica-set) for local commands.
+
+## 4. Seed and start the backend
+
+```bash
+(cd backend && bun run seed)
+(cd backend && bun run dev)
+```
+
+Backend runs at `http://localhost:4000`. OpenAPI spec at `/openapi.json`.
+
+Seed users: `test@example.com` and `admin@example.com`, password `testpassword123`.
+
+## 5. Start the frontend
+
+Keep the backend running while generating the SDK.
+
+```bash
+(cd frontend && bun run sdk)
+(cd frontend && bun run web)
+```
+
+Frontend runs at `http://localhost:8082`. Sign in with `test@example.com` / `testpassword123`.
+
+## Explore the Terreno monorepo examples
+
+If you are contributing to Terreno or want reference implementations inside this repository:
+
+```bash
+git clone https://github.com/TerrenoLabs/terreno.git
+cd terreno && bun run bootstrap
+```
+
+| Service | Port | Command (from repo root) |
+| --- | --- | --- |
+| example-backend | 4000 | `bun run backend:dev` |
+| example-frontend web | 8082 | `EXPO_PUBLIC_API_URL=http://localhost:4000 bun run frontend:web` |
+
+The example backend needs `MONGO_URI` (replica set) and auth secrets — see the repository [AGENTS.md](https://github.com/TerrenoLabs/terreno/blob/master/AGENTS.md) (Cursor Cloud section). Seed with `bun run backend:seed` from the repo root.
+
+## Next steps
+
+- [Create a Terreno app](../how-to/create-a-terreno-app.md) — CLI flags, layout, deploy, MCP write path
+- [How to create a model](../how-to/create-a-model.md) — Model conventions including required field descriptions
+- [How-to guides](../how-to/) — Task-focused guides
+- [Reference](../reference/) — Package and API details