@terreno/mcp 57.9.0 → 57.10.1

package/dist/docs/versioned/next/how-to/observe-llm-calls.md

@@ -0,0 +1,197 @@
+# Observe LLM calls
+
+Register observability on the backend. Registration **is** the feature flag. Then run product iteration in [Develop an AI feature](ai-feature-development.md).
+
+## Register
+
+```typescript
+import {access} from "./access";
+
+const priceMap = parseObservabilityPriceMap(process.env.AI_OBS_PRICE_MAP_JSON);
+
+server.register(new ObservabilityApp({
+  accessControl: access,
+  aiService,
+  plugins: [localPlugin, langfuseAdapter, otelSink].filter(Boolean),
+  control: {
+    datasets: process.env.AI_OBS_DATASETS_PRIMARY ?? "local",
+    experiments: process.env.AI_OBS_EXPERIMENTS_PRIMARY ?? "local",
+    prompts: process.env.AI_OBS_PROMPTS_PRIMARY ?? "local",
+    reviewQueue: "local",
+  },
+  priceMap,
+  sampleRate: Number(process.env.AI_OBS_SAMPLE_RATE ?? 0),
+}));
+```
+
+For an admin playground that accepts a user-supplied provider key, add
+`requestAiServiceFactory: ({apiKey, modelId}) => ...`. The playground route passes
+`x-ai-api-key` to that factory only when the server-wide `aiService` is absent. Never persist or
+log the key. `GET /ai/observability/status` exposes `playgroundAi.source` so admin UIs can block
+**Run once** with a host-specific hint (`request-key`) instead of surfacing the backend **503**
+missing-key title as a server misconfiguration.
+
+Validate the parsed object before registration: each model needs non-negative numeric
+`inputPerMTok` and `outputPerMTok`. The example implementation is
+`example-backend/src/utils/observabilityConfig.ts`; malformed values fail startup with the
+variable name.
+
+Keep existing `LangfuseApp` if you already use Langfuse keys. The Langfuse **adapter** must reuse `getLangfuseClient()` and must not start a second OpenTelemetry SDK.
+
+Boot **fails** if `experiments.primary !== datasets.primary`, if `reviewQueue` is `langfuse`, or if a primary’s plugin is missing.
+
+Pass the same `access` object you use for `TerrenoApp` and `AdminApp` so
+`GET /ai/observability/status` permissions match `/admin/config` screen filtering and every
+observability route enforces RBAC.
+
+### Example consumer roles
+
+The example backend seeds two optional roles in `example-backend/src/rbacRoles.ts`. They are
+**not** assigned automatically — attach them in the admin **Roles** UI or on `User.roles`.
+
+| Role | Grants |
+| --- | --- |
+| `aiObservabilityViewer` | `admin:access` plus `list` and `read` on all six observability resources |
+| `aiObservabilityOperator` | `admin:access` plus every approved action on those resources |
+
+Compose your own roles the same way: grant `admin: ["access"]` for the admin shell, then add
+only the `aiPrompt` / `aiTrace` / `aiReview` / `aiDataset` / `aiEvaluator` / `aiExperiment`
+actions each persona needs. Prompt `folder` and `description` stay consumer-defined metadata —
+Terreno does not ship domain-specific folder names or an in-app approval workflow.
+
+## Env
+
+| Variable | Role |
+| --- | --- |
+| `AI_OBS_PROMPTS_PRIMARY` | `local` or `langfuse` (default `local`) |
+| `AI_OBS_DATASETS_PRIMARY` | default `local` |
+| `AI_OBS_EXPERIMENTS_PRIMARY` | must equal datasets |
+| `AI_OBS_SAMPLE_RATE` | `0`–`1`, default `0` (live evals off) |
+| `AI_OBS_PRICE_MAP_JSON` | `{ "gemini-2.5-flash": { "inputPerMTok": 0.1, "outputPerMTok": 0.4 } }` |
+| `LANGFUSE_PUBLIC_KEY` / `LANGFUSE_SECRET_KEY` | existing Langfuse plugin |
+| `OTEL_EXPORTER_OTLP_TRACES_ENDPOINT` | optional OTLP trace sink |
+
+## App generate calls
+
+Pass identity and the production prompt ref so traces feed the SOP loop:
+
+```typescript
+await aiService.generateText({
+  prompt: text,
+  promptLabel: "production",
+  promptName: "example-summarize",
+  sessionId: req.header("x-ai-session-id"),
+  userId: req.user?.id,
+});
+```
+
+Opt out of tracing with `skipTrace: true`. Sink failures are logged and never fail the generate. `AIRequest` is still written.
+
+## Admin
+
+Phase 1 admin ships **Prompts**, **Traces**, and **Review**. Datasets, Evaluators, and
+Experiments screens arrive in phase 2. **Open in Langfuse** appears in phase 3 when the
+Langfuse plugin is registered. Review is hidden when the local plugin is off.
+
+On **Traces**, **Run multi-stage trace test** (local sink only) runs two schema-backed
+`generateJsonObject` calls, a deterministic tool stage, and a final schema-backed combine
+call, then opens the nested trace. Each LLM span input includes `outputSchema`. It uses the
+server `aiService` when one is configured; otherwise it sends the admin's saved key as
+`x-ai-api-key` to `requestAiServiceFactory`, and the button stays disabled with a hint until
+that key exists.
+
+## Run the example locally
+
+1. Set `MONGO_URI` to a replica set and the auth secrets from
+   `example-backend/.env.example`.
+2. Run `bun run backend:seed`. The idempotent seed creates `examples/example-summarize`
+   v1 with `production` pointing to v1 and candidate v2, plus the human
+   `correctness-human` evaluator, automatic `schema-assert` evaluator, and
+   two-item `example-gold` dataset. It also creates `chat-safety-synthetic` (12 proofread
+   two-person chats), `chat-safety-screen` v1 and v2, and the `chat-safety-agreement`
+   evaluator. On a database that already booted, run Admin → Scripts →
+   `seedChatSafetyDataset` as a wet run, or
+   `cd example-backend && bun run script seedChatSafetyDataset --wet`.
+3. Start `bun run backend:dev` and `bun run frontend:web`, then sign in as the seeded
+   admin.
+4. Open **AI Observability → Prompts** to inspect or save a new immutable version. Move
+   `production` explicitly; the seed never overwrites an existing label.
+5. Configure `GOOGLE_VERTEX_PROJECT` or `GEMINI_API_KEY`, then sign in with Better Auth
+   and call the seeded route:
+
+   ```bash
+   API=http://localhost:4000
+   COOKIE_JAR="$(mktemp)"
+   curl -sS -c "$COOKIE_JAR" \
+     -H "Content-Type: application/json" \
+     -d '{"email":"admin@example.com","password":"testpassword123"}' \
+     "$API/api/auth/sign-in/email" >/dev/null
+
+   curl -X POST "$API/ai/example-summarize" \
+     -b "$COOKIE_JAR" \
+     -H "Content-Type: application/json" \
+     -H "x-ai-session-id: docs-walkthrough" \
+     -d '{"text":"Terreno records local nested traces for every AI call."}'
+   ```
+
+6. Open its `example-summarize` trace, send it to Review, score **correct**, and submit
+   until the queue reports **Queue clear**.
+
+The example app also calls this route from the UI: **Todos → Summarize** sends the user's todo
+list, so each press produces a trace with the production prompt reference, `userId`, and a
+per-app-run `x-ai-session-id`. Without server credentials the screen forwards the Gemini key
+saved in **Profile** as `x-ai-api-key`; otherwise the route answers **503**.
+
+The route sends `text` as the user prompt. The labelled registry version supplies the
+system prompt and the route uses `TemperaturePresets.LOW`.
+
+The example registers the local plugin even when no Langfuse keys exist. Set
+`AI_OBS_PRICE_MAP_JSON` to calculate USD cost; unlisted models show tokens with no cost,
+never `$0`.
+
+## Create a prompt and pin production
+
+Admin-only routes live under `/ai/observability`. Create the prompt in a folder, save later edits as `vN+1` (v1 stays immutable), then move `production` — the response names the outgoing version for the confirm modal.
+
+```bash
+# 1. Create v1 in a folder
+curl -X POST "$API/ai/observability/prompts" \
+  -H "Authorization: Bearer $ADMIN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"folder":"examples","name":"example-summarize","system":"Summarize {{text}}","template":"{{text}}","type":"text"}'
+
+# 2. Pin production (first pin has no outgoing version)
+curl -X POST "$API/ai/observability/prompts/example-summarize/labels" \
+  -H "Authorization: Bearer $ADMIN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"label":"production","version":1}'
+
+# 3. Save as v2, then move production (response.outgoingVersion is 1)
+curl -X POST "$API/ai/observability/prompts/example-summarize/versions" \
+  -H "Authorization: Bearer $ADMIN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"system":"Summarize {{text}} in one sentence","template":"{{text}}","type":"text"}'
+curl -X POST "$API/ai/observability/prompts/example-summarize/labels" \
+  -H "Authorization: Bearer $ADMIN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"label":"production","version":2}'
+```
+
+`GET /ai/observability/prompts?folder=examples&search=sum&include=usage7d` lists folder matches with 7-day call/cost rollups. A prompt with no `production` label returns `production: "—"`. Playground `POST /ai/observability/prompts/:name/playground` compiles `{{var}}`, runs one `AIService` call, and does not create a version.
+
+The example backend registers `requestAiServiceFactory` but no server `aiService`, so status reports
+`playgroundAi.source: "request-key"`. In **AI Observability → Prompts → Playground**, save a Gemini
+API key on **Profile** first; the admin UI blocks **Run once** with that hint until the key loads
+from storage and is sent as `x-ai-api-key`. **Traces → Run multi-stage trace test** uses the same
+saved key and is blocked the same way.
+
+## Review a trace
+
+1. Use the seeded human `correctness-human` evaluator. If you skipped `bun run backend:seed`,
+   install it once with `POST /ai/observability/evaluators/templates/correctness-human`; a
+   duplicate install returns 409.
+2. Enqueue traces: `POST /ai/observability/traces/review` with `{evaluatorId, traceIds, reason: "manual"}`.
+3. List oldest-first: `GET /ai/observability/review?status=pending` (response includes per-status counts).
+4. Open an item: `GET /ai/observability/review/:id` returns evaluator dimensions plus `given` / `wrote` panels (variable labels when present, otherwise raw keys).
+5. `POST /ai/observability/review/:id` with `action: "submit"` writes dimension-keyed scores through every ScoreSink and marks the item `done` (it leaves the pending list). `skip` and `assign` move status to `skipped` / `in_progress`.
+

package/dist/docs/versioned/next/how-to/run-tests-locally.md

@@ -48,6 +48,7 @@ bun test --only-failures <path>
 | --- | --- |
 | `api-ci` | `bun run api:test` |
 | `ai-ci` | `cd ai && bun run test:ci` |
+| `blocks-ci` | `bun run blocks:test` |
 | `ui-ci` | `bun run ui:test` |
 | `rtk-ci` | `bun run rtk:test` |
 | `syncdb-ci` | `bun run syncdb:test` |
@@ -81,7 +82,7 @@ MONGO_URI="mongodb://127.0.0.1:27017/terreno-e2e?replicaSet=rs0" \
 | `e2e` `auth` | `login` `signup` `consents` `forgot-password` `reset-password` `verify-email` |
 | `e2e` `app` | `todos` `profile` `realtime` `ai-chat` `pdf` |
 | `e2e` `admin-core` | `admin` `admin-home` `admin-form` `admin-todo-crud` `admin-title-update-depth` |
-| `e2e` `admin-table` | `admin-table-search-filter` `admin-table-bulk-actions` `admin-custom-screens` `admin-comms-back` |
+| `e2e` `admin-table` | `admin-table-search-filter` `admin-table-bulk-actions` `admin-custom-screens` `admin-comms-back` `harness-approvals` `harness-intake` |
 | `e2e` `syncdb` | `syncdb-load-delta` `syncdb-offline` `syncdb-conflicts` `syncdb-storage` `syncdb-chaos` |
 
 Full local suite: `bun run frontend:e2e`. Nightly load (`e2e-load` / `syncdb-loadlab`) is

package/dist/docs/versioned/next/how-to/ship-a-new-task-version.md

@@ -0,0 +1,76 @@
+# Ship a new task version
+
+Change a durable harness task without breaking runs that already started on the old code.
+Rules: [Versioning](../reference/ai-harness.md#versioning). Why:
+[Version pinning](../explanation/durable-agent-harness.md#version-pinning).
+
+## When to bump `version`
+
+Bump it on any change an in-flight task could notice:
+
+- A phase is renamed, removed, or added between existing phases.
+- The shape of `state`, `input`, or the result changes.
+- A phase's side effects or `replay` policy change.
+
+A fix that keeps the same phases, state shape, and side effects can ship under the same
+version.
+
+## Steps
+
+1. Copy the definition and bump `version`. Keep the old one unchanged.
+
+   ```typescript
+   export const intakeV1 = defineTask<IntakeInput, IntakeStateV1, IntakeResult>({
+     name: "clinic.intake",
+     version: 1,
+     // ...the old phases, untouched
+   });
+
+   export const intakeV2 = defineTask<IntakeInput, IntakeStateV2, IntakeResult>({
+     name: "clinic.intake",
+     version: 2,
+     // ...the new phases
+   });
+   ```
+
+2. Register both versions. Point new work at v2.
+
+   ```typescript
+   const harness = await Harness.open({registry: [intakeV1, intakeV2]});
+   await harness.start();
+
+   await harness.createTask(intakeV2, {patientId});
+   ```
+
+   Tasks created on v1 finish on v1's phases and v1's `abort` handler. New tasks run on
+   v2. Parent tasks that call `rt.createTask(intakeV1, ...)` keep making v1 children
+   until you change that call.
+
+3. Deploy. Every process that runs the harness needs the same registry.
+
+4. Wait until no v1 task is left in flight. Check in a Mongo shell:
+
+   ```javascript
+   db.harnesstasks.countDocuments({
+     name: "clinic.intake",
+     version: 1,
+     status: {$in: ["pending", "running", "waiting", "interrupted"]},
+   });
+   ```
+
+   Resolve stuck `interrupted` v1 tasks with `harness.resolveInterrupted`, or stop them
+   with `harness.abort`.
+
+5. When the count is 0, remove `intakeV1` from the registry and deploy again.
+
+## If you remove a version too early
+
+`Harness.start()` throws an `APIError` with code `harness-config-invalid` and claims
+nothing. Its `detail` reads:
+
+```text
+Harness.start: in-flight tasks use task versions this registry does not register: clinic.intake@1 (4 tasks). Register those definitions (keep old versions until their tasks finish) or resolve the tasks first.
+```
+
+Fix: add the old definition back to `registry` and redeploy. Its tasks resume from their
+last checkpoint.

package/dist/docs/versioned/next/how-to/upgrade-terreno.md

@@ -38,7 +38,7 @@ Maintainers watching the next SDK beta use `track-upstream-expo` / [`track-upstr
 
 ### Backend vs frontend packages
 
-From `publish-on-tag.yml`: **backend** `api`, `test`, `admin-backend`, `ai`, `api-health`, `comms`, `feature-flags`, `mcp`. **frontend** `ui`, `rtk`, `syncdb`, `admin-frontend`, `admin-spa`.
+From `publish-on-tag.yml`: **backend** `api`, `test`, `admin-backend`, `blocks`, `ai`, `api-health`, `comms`, `feature-flags`, `mcp`. **frontend** `ui`, `rtk`, `syncdb`, `admin-frontend`, `admin-spa`. `blocks` holds contracts that both `ai` and `ui` depend on, so it moves with the backend step.
 
 ## 4. RTK → syncdb is not a version bump