@@ -22,6 +22,7 @@ npm test
 ## Public API
 
 - `AgilityApiClient` - top-level SDK client (`meta`, `data` for rest-1, `assets` for api/asset, `localize` for loc-2, `schema`), and the services on it:
+  - `queryV1.query` / `queryV1.queries` - reads through query.v1: the same query bodies as `assets.query`, rows translated the same way, but the endpoint is read-only and answers with rows alone, so it is the safe place to explore a query
   - `resolve.numbers` / `resolve.ref` - asset numbers and OID tokens to assets, in one request
   - `assets.sendGuarded` - a batch as a compare-and-set on each asset's ChangeDate; changed assets come back `skipped`
   - `batch.run` / `batch.reversible` - preview, snapshot, (confirm,) send guarded, and a receipt with a reverse batch
 
@@ -134,8 +134,11 @@ Found by reading the code above against `src/`:
    return failures typed (it does) and offer `throwOnFailure` producing an
    `ApiError` subclass carrying `sourceCommandIndex` per failure.
 5. **Missing wire modules** named in spec 000 but not built: `loc2` (both
-   consumers), `queryV1` (slack's main read path — or migrate slack to
-   api/asset queries instead; see B3), `attachment`/`EmbeddedImage` (rooms).
+   consumers), `queryV1` (slack's main read path), `attachment`/
+   `EmbeddedImage` (rooms). `queryV1` is built (2026-11-06) and kept on
+   purpose even though api/asset can run the same queries: query.v1 is
+   read-only by construction and answers with rows alone, so it is the safe,
+   simple place to explore a query (and its requests may be retried).
 6. **rest-1 writes — not needed.** Settled from the docs
    (apis/asset-bulk/updating-assets, creating-assets): api/asset carries a
    change comment as `ChangeComment` inside an update's `update` node, or
@@ -166,11 +169,11 @@ moves.
 - **A2 Command-failure ergonomics** — gap 4: `CommandFailureError`,
   `send(..., { throwOnFailure })`, `sourceCommandIndex` preserved.
 - **A3 loc2 wire module** — gap 5, smallest consumer-shared endpoint.
-- **A4 Slack-shaped reads** — decide queryV1-module vs. api/asset-only by
-  porting `lib/queries.js` bodies into integration tests against the dev
-  instance; whichever reads translate cleanly wins. (No `rest1` writes:
-  gap 6 is settled.) If the bot's S2 lands first, this is only the
-  translation check.
+- **A4 Slack-shaped reads** — done: a `queryV1` wire module and
+  `client.queryV1.query`/`queries`, rows translated through the same
+  schema as api/asset (the integration tier proves both endpoints give the
+  same rows). The bot keeps query.v1 for its reads. (No `rest1` writes:
+  gap 6 is settled.)
 - **A5 Services port** — gap 7. `services/resolve` first (gh-v1's
   `resolveAssets`; slack's `queryForAssetByNumber` needs it too), then
   `services/guard` (`guarded()`), then `services/batch` (preview, describe,
@@ -200,12 +203,12 @@ bodies in `lib/queries.js`), api/asset (`bulk`, `createWithNumber`,
 
 - **B0 Pre-SDK migrations in the bot's own session** (S1–S4 under
   *Consumer migrations* below). They shrink what the SDK must support:
-  once S2 moves reads to api/asset, the SDK needs no `queryV1` module and
-  A4 collapses to "verify the bot's query bodies translate" — and A4's
-  rest-1-writes question is closed (gap 6).
+  once S2 pages every read, B1's reads move to `client.queryV1` as they
+  are; A4's rest-1-writes question is closed (gap 6).
 - **B1 Client swap.** `createV1Client` keeps its surface; inside, one
-  `AgilityApiClient({ baseUrl: url, accessToken })`. `query`/`bulk`/`exec`
-  → `client.assets.*`; `fetchLocalizations` → `loc2` (A3).
+  `AgilityApiClient({ baseUrl: url, accessToken })`. `query` →
+  `client.queryV1.queries`; `bulk`/`exec` → `client.assets.*`;
+  `fetchLocalizations` → `loc2` (A3).
   `test/v1-service.test.js` stays the parity proof.
 - **B2 Delete the decodes.** Rows arrive translated: `"True"` comparisons,
   `IsKanban.toLowerCase()`, `toMomentless` and positional
@@ -225,8 +228,9 @@ instance: same requests out, same rows in.
 
 - **C1 Seam swap.** Re-implement `lib/api.js`'s exported surface (`query`,
   `queries`, `asset`, `attempt`, `rest`, `meta`, `localize`, `raw`, `blob`,
-  `permalink`) on `AgilityApiClient`, `require`d from the CJS build. Callers
-  unchanged; all 99 tests green. Needs A2, A3 (loc2), gap 9.
+  `permalink`) on the SDK, imported from a vendored copy of its ESM build
+  (gh installs extensions without npm; see plan 002, *Dependency*). Callers
+  unchanged; all tests green. Needs A2, A3 (loc2).
 - **C2 Delete the decodes.** With rows translated by the schema, `isTrue`
   and `oidOf` drop out of `meta.js`, `resolve.js`, `board.js`; `momentless`
   comes from the SDK.
@@ -300,10 +304,9 @@ ESM" here means finishing the edges:
 - **S1 Finish ESM tooling.** `jsconfig.json` still says `module: commonJS`,
   `target: es6`; set `NodeNext`/`ES2022` and turn on `checkJs` so editors
   and `tsc --noEmit` see the real module system (and, later, SDK types).
-- **S2 query.v1 → api/asset reads.** Same query bodies; the endpoint is
-  `api/asset` and results come from `queryResult.results` (positional, one
-  array per query). Add an explicit `page` to every body. Removes the
-  SDK's need for a `queryV1` module.
+- **S2 Page every read.** Reads stay on query.v1 (read-only, simplest
+  responses). Add an explicit `page` to every body in `lib/queries.js`,
+  as the SDK's `queryV1` refuses an unpaged query.
 - **S3 Close Bolt Phase 2.** Drop `botkit`, `botbuilder-*`,
   `botkit-plugin-cms` and `axios` from `package.json` once nothing imports
   them; delete `lib/v1api.js` and the `plugins/` Botkit modules that
@@ -354,12 +357,12 @@ ESM" here means finishing the edges:
 - [x] A1 transport: credentials, onUnauthorized (idempotent requests resent once; writes never), signal (`RequestOptions.signal`, `client.withSignal`)
 - [x] A2 command-failure error (`CommandFailureError`, `throwOnFailure`, reads always throw)
 - [x] A3 loc2 (`loc2.localize`, `client.localize`; GET, POST past 2000 encoded chars, culture via Accept-Language)
-- [ ] A4 slack-shaped reads (queryV1 or api/asset)
+- [x] A4 slack-shaped reads — `queryV1` wire module, `client.queryV1.query`/`queries` (read-only, idempotent, same row translation as api/asset)
 - [x] A5 services port from gh-v1 — `resolve` (`client.resolve.numbers`/`.ref`), `guard` (`client.assets.sendGuarded`), `batch` (`client.batch.run`/`.reversible`)
 - [ ] A6 CI + prerelease publish — CI (`.github/workflows/ci.yml`: verify on Node 20/22/24, `npm pack` artifact) and git-SHA installs (`prepare`) done; publish waits on the registry decision
 - [ ] G0–G3 gh-v1 migrations (gh-v1 session)
 - [ ] S1–S4 slack-bot migrations (slack-bot session, `bolt-migration`)
 - [ ] B1–B4 slack bot SDK swap — brief: `plans/003-adopt-sdk-k6i-agility-slack-bot.md`
 - [x] C0 gh-v1 surveyed (2026-11-04)
-- [ ] C1–C4 gh-v1 — brief: `plans/002-adopt-sdk-gh-v1.md`
+- [ ] C1–C4 gh-v1 — brief: `plans/002-adopt-sdk-gh-v1.md`; C1 in versionone/gh-v1#33 (SDK vendored)
 - [ ] D1–D4 agility-rooms
@@ -25,16 +25,27 @@ output stay the same; the code that duplicates the SDK goes.
   in `lib/api.js`, `meta.js`, `resolve.js`, `board.js`, `batch.js`, and
   `commands/apply.js`, `bulk.js`, `board.js`.
 
-## Dependency
+## Dependency: vendored
 
-No registry package yet. Add the first runtime dependency as a pinned commit:
-`"@versionone/sdk": "github:versionone/VersionOne.SDK.JavaScript#<merge sha>"`.
-The SDK's `prepare` script builds `dist` on install. gh-v1 is ESM now, so use
-`import` (the ESM build), not `require`.
+gh installs a script extension with a plain `git clone` and never runs npm,
+so gh-v1 cannot rely on `node_modules` at run time. The SDK is still a
+runtime dependency, but its code ships inside the repo:
+
+- `vendor/versionone-sdk/` holds the SDK's ESM build (`index.js`, which
+  imports nothing) and types (`index.d.ts`), plus `SOURCE.json` naming
+  the commit they came from. `lib/api.js` imports from there.
+- `package.json` lists `@versionone/sdk` under `devDependencies`, pinned
+  to a full commit, so the lockfile records exactly what was vendored.
+- To move the pin: change it in `package.json`, then
+  `npm install && npm run vendor:sdk`, and commit the copy.
+  `test/vendor.test.js` fails if the copy, `SOURCE.json` and the pin
+  disagree.
+- Publishing the SDK to a registry (plan 001, *Decisions needed*) changes
+  only where the pin points; the copy stays.
 
 ## Steps, one PR each, on branch `sdk-adoption`
 
-1. **C1 Seam swap** (`lib/api.js`). Keep `api(cfg)` and the `Api` typedef.
+1. **C1 Seam swap** (`lib/api.js`) — done in versionone/gh-v1#33. Keep `api(cfg)` and the `Api` typedef.
    Inside, build one `AgilityApiClient({ baseUrl: cfg.v1Url, accessToken })`.
    Map: `query`/`queries` → `client.assets.query`/`queries`;
    `asset` → `client.assets.send(cmds, { throwOnFailure: true })`
 
@@ -21,7 +21,7 @@ hand-decoding and rest-1 write path go.
   `fetchLocalizations` (loc-2), and `queryFor*` helpers built on
   `lib/queries.js`. `test/v1-service.test.js` stubs global `fetch`.
 - **S1–S4 are not done yet**: `jsconfig.json` still says `commonJS`/`es6`;
-  reads still post to `query.v1`; `axios`, `botkit*`, `botbuilder-*` are
+  query bodies have no explicit page; `axios`, `botkit*`, `botbuilder-*` are
   still dependencies and `lib/v1api.js` still exists. These come first.
 
 ## Dependency
@@ -34,12 +34,12 @@ The SDK's `prepare` script builds `dist` on install. The bot is ESM, so
 ## Steps, one PR each
 
 0. **S1–S4 first**, as written in `plans/handoffs/k6i-agility-slack-bot.md`.
-   B1 needs at least S2 (reads on api/asset) and S3 (axios and Botkit gone).
+   B1 needs at least S2 (every read paged) and S3 (axios and Botkit gone).
    If the bot's own session already merged them, skip.
 1. **B1 Client swap** (`services/v1.js`). Inside `createV1Client`, build one
    `new AgilityApiClient({ baseUrl: url, accessToken })`. `query` →
-   `client.assets.queries` (positional; a failed query now throws instead
-   of shifting later results); `bulk`/`exec`/`createWithNumber` →
+   `client.queryV1.queries` (the bot's reads stay on query.v1: read-only,
+   simplest responses; a refused query throws with the server's message); `bulk`/`exec`/`createWithNumber` →
    `client.assets.send`; `fetchLocalizations` → `client.localize`;
    `findAssets` → `client.data.list("BaseAsset", { find, findin, where,
    sort, page, deleted: false })`. Map `ApiError` with `status === 401`
@@ -72,8 +72,9 @@ The SDK's `prepare` script builds `dist` on install. The bot is ESM, so
 ## Done when
 
 - `npm test` passes on each PR; S1's `npm run typecheck` passes.
-- `services/v1.js` makes no `fetch` call of its own; nothing posts to
-  `query.v1` or writes through rest-1.
+- `services/v1.js` makes no `fetch` call of its own; reads go through
+  `client.queryV1` (or `client.data` for `find`), and nothing writes through
+  rest-1.
 - No string-boolean comparisons or local `toMomentless` remain.
 - Sandbox-workspace smoke passes: asset mention, `/agility find`,
   backlog-item status change, webhook notification.
 
@@ -21,16 +21,12 @@ no `require`). `jsconfig.json` still declares `module: commonJS`,
 `npm run typecheck` (`tsc -p jsconfig.json`; `typescript` is already a
 devDependency). Start non-strict; fix or annotate what it finds.
 
-**S2 — Reads from query.v1 to api/asset.** In `services/v1.js`, `query()`
-posts to `query.v1`. Post the same bodies to `api/asset` instead and read
-`queryResult.results` (one array per query, positional, an empty array
-for a query that matched nothing) so callers still receive an array of
-result arrays. Add an explicit `page: { size, start }` to every body in
-`lib/queries.js` (api/asset reads should never be unbounded). Note that
-api/asset returns every scalar as a string (`"True"`, `"5"`) and a
-single relation selected with a sub-select as a one-element list — check
-each call site's comparisons against a real response. Update
-`test/v1-service.test.js`.
+**S2 — Page every read.** Reads stay on `query.v1`: it is read-only and
+answers with rows alone, and the SDK has a `queryV1` module for it. Add an
+explicit `page: { size, start }` to every body in `lib/queries.js` (the
+SDK refuses an unpaged query; no read should be unbounded), choosing sizes
+that match what each caller actually shows. Update
+`test/v1-service.test.js` to assert the pages sent.
 
 **S3 — Close Bolt Phase 2.** AGENTS.md says axios is dropped in Phase 2.
 Once nothing imports them, remove `axios`, `botkit`,
 
@@ -132,9 +132,11 @@ expect-zero control beside every "it returned rows".
 
 ## 6. Politeness is in the layer
 
-No read is unbounded: `assetApi.queries` refuses a query without a page,
+No read is unbounded: `assetApi.queries` and `queryV1.queries` refuse a
+query without a page,
 and the rest-1 module will refuse a Data read without one. The transport
-retries only idempotent requests (GET by default), only on a network
+retries only idempotent requests (GET by default, and every query.v1
+POST, since that endpoint cannot write), only on a network
 failure or a 5xx, never a 4xx; a batch write is never retried unless the
 caller has made it idempotent with a guard and says so.
 
@@ -156,7 +158,10 @@ always throw: a query the server rejected must not come back as `[]`.
 The server drops a failed query's slot from `queryResult.results` (later
 results shift up); `send` restores it as `null` by `sourceCommandIndex`, so
 positions always match the queries sent, and refuses to guess when the
-counts disagree.
+counts disagree. If a failure carries no sourceCommandIndex it may be any
+of the queries, so every query slot is null and the failures are the
+answer (thrown for reads and with throwOnFailure); a disagreement with no
+such failure is an error of its own.
 
 The guard (`services/guard`) is what makes a write batch safe to resend:
 each per-asset command of the server's projection is sent scoped to its
 
@@ -10,6 +10,7 @@ import type { Metamodel } from "./types/metamodel";
 import * as assetApi from "./wire/assetApi";
 import * as loc2 from "./wire/loc2";
 import * as meta from "./wire/meta";
+import * as queryV1 from "./wire/queryV1";
 import * as rest1 from "./wire/rest1";
 
 export interface ClientOptions {
@@ -113,6 +114,22 @@ export class AgilityApiClient {
     history: <Row extends Asset = Asset>(target: string, params: rest1.ReadParams) => rest1.history<Row>(this.transport, target, params)
   };
 
+  /**
+   * query.v1: reads only, the same query bodies as assets.query, rows
+   * translated the same way. The endpoint cannot write, so it is the safe
+   * place to explore a query; its requests may be retried like a GET.
+   */
+  readonly queryV1 = {
+    query: async <Row extends Asset = Asset>(q: assetApi.QueryBody) => {
+      await this.schema.ensureFor([q]);
+      return queryV1.query<Row>(this.transport, q, { lookup: this.schema });
+    },
+    queries: async <Row extends Asset = Asset>(qs: assetApi.QueryBody[]) => {
+      await this.schema.ensureFor(qs);
+      return queryV1.queries<Row>(this.transport, qs, { lookup: this.schema });
+    }
+  };
+
   /** api/asset: batches and queries, rows translated through the schema. */
   readonly assets = {
     send: async <Row extends Asset = Asset>(commands: assetApi.Command[], options?: Omit<assetApi.SendOptions, "lookup">) => {
 
@@ -15,7 +15,7 @@ export type { AssetType, AttributeDefinition, AttributeLookup, AttributeType, At
 export type * from "./types/assets/index";
 
 // The wire layer: one module per endpoint.
-export { assetApi, rest1, meta, loc2 } from "./wire";
+export { assetApi, rest1, meta, loc2, queryV1 } from "./wire";
 export * as select from "./wire/select";
 export type { AssetApiResponse, Command, CommandBody, CommandFailureDetail, Outcome, CommandValue, CreateBody, ExecuteBody, QueryBody, UpdateBody, ReadOptions, RowContext, SendOptions } from "./wire/assetApi";
 export { CommandFailureError } from "./wire/assetApi";
 
@@ -283,12 +283,13 @@ function queriesIn(commands: Command[]): { body: QueryBody; index: number }[] {
  * The server's results, one slot per query again. It sends a slot for every
  * query that ran (an empty match included) and none for a query that
  * failed; a failure names its command by sourceCommandIndex, so the failed
- * queries are known and the rest fill in order. If the counts disagree the
- * positions cannot be trusted, and that is an error rather than a guess.
+ * queries are known and the rest fill in order. Null when the counts
+ * disagree: the positions cannot be trusted, and the caller decides what
+ * that means rather than this guessing.
  */
-function realign<T>(sent: { index: number }[], results: T[], failed: Set<number>): (T | null)[] {
+function realign<T>(sent: { index: number }[], results: T[], failed: Set<number>): (T | null)[] | null {
   const expected = sent.filter((q) => !failed.has(q.index)).length;
-  if (results.length !== expected) throw new Error(`api/asset returned ${results.length} query results for ${expected} queries that did not fail; cannot tell which result belongs to which query`);
+  if (results.length !== expected) return null;
   let next = 0;
   return sent.map((q) => (failed.has(q.index) ? null : results[next++]!));
 }
@@ -311,13 +312,24 @@ export async function send<Row extends Asset = Asset>(transport: Transport, comm
   // A preview answers with the projection, not an envelope.
   if (options.previewOnly) return response;
   const queries = queriesIn(commands);
-  const failed = new Set(failuresOf(response).flatMap((f) => (f.index === null ? [] : [f.index])));
-  const aligned = realign(queries, response.queryResult?.results ?? [], failed);
+  const failures = failuresOf(response);
+  const failed = new Set(failures.flatMap((f) => (f.index === null ? [] : [f.index])));
+  const results = response.queryResult?.results ?? [];
+  let aligned = realign(queries, results, failed);
+  if (!aligned) {
+    // A failure without a sourceCommandIndex may be one of the queries, so
+    // no row can be placed with confidence: every slot is null, never a
+    // misplaced row, and the failures themselves are the answer (thrown
+    // with throwOnFailure, as every read asks). With no such failure, the
+    // server's answer is inconsistent, and that is an error of its own.
+    if (!failures.some((f) => f.index === null)) throw new Error(`api/asset returned ${results.length} query results for ${queries.filter((q) => !failed.has(q.index)).length} queries that did not fail; cannot tell which result belongs to which query`);
+    aligned = queries.map(() => null);
+  }
   const contextOf = (i: number): RowContext | undefined => {
     const type = options.lookup ? fromTypeOf(queries[i]!.body.from, commands) : null;
     return type ? { lookup: options.lookup!, type } : undefined;
   };
-  const translated: AssetApiResponse<Row> = { ...response, queryResult: { ...response.queryResult, results: aligned.map((rows, i) => rows && rows.map((r) => fromWireRow<Row>(r as Record<string, unknown>, contextOf(i)))) } };
+  const translated: AssetApiResponse<Row> = { ...response, queryResult: { ...response.queryResult, results: aligned!.map((rows, i) => rows && rows.map((r) => fromWireRow<Row>(r as Record<string, unknown>, contextOf(i)))) } };
   if (options.throwOnFailure && translated.commandFailures?.count) throw new CommandFailureError(translated);
   return translated;
 }
 
@@ -7,6 +7,7 @@ export * as assetApi from "./assetApi";
 export * as rest1 from "./rest1";
 export * as meta from "./meta";
 export * as loc2 from "./loc2";
+export * as queryV1 from "./queryV1";
 export * as select from "./select";
 export type { AssetDocument, AssetsDocument, ReadParams, ReadResult } from "./rest1";
 export type { AssetApiResponse, Command, CommandBody, CommandValue, CreateBody, ExecuteBody, QueryBody, UpdateBody, CommandFailure, CommandError, OidTokens, SendOptions } from "./assetApi";