improvement(search): accept several same-kind native queries per account and cap oversized read limits

Author: waleedlatif1

apps/sim/lib/api/contracts/mothership-assistant-tools.test.ts

@@ -57,13 +57,13 @@ describe('Assistant execution contracts', () => {
     expect(accepts([trip, { ...trip, accountId: 'other' }])).toBe(true)
   })
 
-  it('takes one GitHub or GitLab query per account and kind', () => {
+  it('runs several same-kind GitHub queries on one account but refuses kind fan-out overlap', () => {
     const accepts = (nativeQueries: Record<string, string>[]) =>
       searchWorkspaceInputSchema.safeParse({ query: 'launch', nativeQueries }).success
     const github = { provider: 'github', accountId: 'account', query: 'repo:org/repo launch' }
     const kinds = ['issues', 'commits', 'code', 'repositories'].map((kind) => ({ ...github, kind }))
     expect(accepts(kinds)).toBe(true)
-    expect(accepts([kinds[0]!, { ...kinds[0]!, query: 'repo:org/repo deploy' }])).toBe(false)
+    expect(accepts([kinds[0]!, { ...kinds[0]!, query: 'repo:org/repo deploy' }])).toBe(true)
     expect(accepts([kinds[0]!, github])).toBe(false)
     expect(
       searchWorkspaceInputSchema.safeParse({ nativeQueries: [{ provider: 'github', query: '' }] })

apps/sim/lib/api/contracts/mothership-assistant-tools.ts

@@ -102,10 +102,10 @@ export const nativeSearchQueriesSchema = z
         addIssue('Duplicate native query.')
       else if (
         hasSearchKinds(query.provider) &&
-        earlier.some((previous) => !previous.kind || !query.kind || previous.kind === query.kind)
+        earlier.some((previous) => !previous.kind || !query.kind)
       )
         addIssue(
-          'Send one query per account and kind for GitHub, GitLab, or HubSpot. Use provider-supported operators for alternatives, or send another call.'
+          'A GitHub, GitLab, or HubSpot query without a kind already searches every kind; give each query on this account a kind.'
         )
       else if (busiestAccountLoad(earlier) >= MAX_NATIVE_QUERIES_PER_ACCOUNT)
         addIssue(
@@ -246,11 +246,12 @@ export const readDocumentInputSchema = z.object({
     .min(1)
     .max(4000)
     .describe('Canonical document ID returned by search or selected document context.'),
+  /** A larger request is capped rather than refused: the server returns at most 8 chunks anyway. */
   limit: z
-    .number()
-    .int()
-    .min(1)
-    .max(8)
+    .preprocess(
+      (limit) => (typeof limit === 'number' && limit > 8 ? 8 : limit),
+      z.number().int().min(1).max(8)
+    )
     .default(3)
     .describe(
       'Maximum number of chunks, from 1 to 8 (default 3); the server may return fewer to fit its text budget. Follow next for more context.'

apps/sim/lib/mothership/generated/sim-assistant-tools.generated.ts

@@ -119,10 +119,10 @@ export const nativeSearchQueriesSchema = z
         addIssue('Duplicate native query.')
       else if (
         hasSearchKinds(query.provider) &&
-        earlier.some((previous) => !previous.kind || !query.kind || previous.kind === query.kind)
+        earlier.some((previous) => !previous.kind || !query.kind)
       )
         addIssue(
-          'Send one query per account and kind for GitHub, GitLab, or HubSpot. Use provider-supported operators for alternatives, or send another call.'
+          'A GitHub, GitLab, or HubSpot query without a kind already searches every kind; give each query on this account a kind.'
         )
       else if (busiestAccountLoad(earlier) >= MAX_NATIVE_QUERIES_PER_ACCOUNT)
         addIssue(
@@ -263,11 +263,12 @@ export const readDocumentInputSchema = z.object({
     .min(1)
     .max(4000)
     .describe('Canonical document ID returned by search or selected document context.'),
+  /** A larger request is capped rather than refused: the server returns at most 8 chunks anyway. */
   limit: z
-    .number()
-    .int()
-    .min(1)
-    .max(8)
+    .preprocess(
+      (limit) => (typeof limit === 'number' && limit > 8 ? 8 : limit),
+      z.number().int().min(1).max(8)
+    )
     .default(3)
     .describe(
       'Maximum number of chunks, from 1 to 8 (default 3); the server may return fewer to fit its text budget. Follow next for more context.'

apps/sim/lib/mothership/generated/tool-catalog-v1.ts

@@ -5196,7 +5196,6 @@ export const ReadDocument: ToolCatalogEntry = {
         description: 'Canonical document ID returned by search or selected document context.',
       },
       limit: {
-        default: 3,
         description:
           'Maximum number of chunks, from 1 to 8 (default 3); the server may return fewer to fit its text budget. Follow next for more context.',
         type: 'integer',

apps/sim/lib/mothership/generated/tool-schemas-v1.ts

@@ -5184,7 +5184,6 @@ export const TOOL_RUNTIME_SCHEMAS: Record<string, ToolRuntimeSchemaEntry> = {
           description: 'Canonical document ID returned by search or selected document context.',
         },
         limit: {
-          default: 3,
           description:
             'Maximum number of chunks, from 1 to 8 (default 3); the server may return fewer to fit its text budget. Follow next for more context.',
           type: 'integer',

apps/sim/lib/mothership/tools/server/knowledge/workspace-search.test.ts

@@ -376,7 +376,7 @@ describe('Assistant retrieval tools', () => {
     expect(mocks.search).not.toHaveBeenCalled()
     expect(mocks.read).not.toHaveBeenCalled()
   })
-  it('reads a selected document through the shared use case and rejects unbounded pages', async () => {
+  it('reads a selected document through the shared use case and caps oversized pages', async () => {
     expect(
       await readDocumentServerTool.execute({ documentId: 'doc', startChunkIndex: 20 }, context)
     ).toMatchObject({ success: true })
@@ -392,7 +392,9 @@ describe('Assistant retrieval tools', () => {
     )
     expect(
       await readDocumentServerTool.execute({ documentId: 'doc', limit: 9 }, context)
-    ).toMatchObject({ success: false })
-    expect(mocks.read).toHaveBeenCalledOnce()
+    ).toMatchObject({ success: true })
+    expect(mocks.read).toHaveBeenLastCalledWith(
+      expect.objectContaining({ input: expect.objectContaining({ limit: 8 }) })
+    )
   })
 })