fix(search): add GitHub commit search and accept native-only live queries

Author: waleedlatif1

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

@@ -73,5 +73,12 @@ describe('Assistant execution contracts', () => {
       topK: 20,
     })
     expect(TOOL_CATALOG.read_document?.parameters).toMatchObject({ required: ['documentId'] })
+    const nativeQueries = [{ provider: 'github', query: 'author:@me', kind: 'commits' }]
+    expect(searchWorkspaceInputSchema.parse({ nativeQueries })).toMatchObject({ query: '' })
+    expect(searchWorkspaceInputSchema.safeParse({}).success).toBe(false)
+    expect(
+      searchWorkspaceInputSchema.safeParse({ nativeQueries: [{ provider: 'github', query: '' }] })
+        .success
+    ).toBe(false)
   })
 })

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

@@ -10,7 +10,9 @@ export const nativeSearchQuerySchema = z
     provider: liveSearchProviderSchema,
     query: z.string().trim().max(2000),
     accountId: z.string().min(1).max(200).optional(),
-    kind: z.enum(['issues', 'code', 'repositories', 'merge_requests', 'wiki']).optional(),
+    kind: z
+      .enum(['issues', 'code', 'repositories', 'commits', 'merge_requests', 'wiki'])
+      .optional(),
     project: z.string().min(1).max(300).optional(),
     cursor: z.string().max(4000).optional(),
     termClauses: z.array(z.string().max(500)).max(10).optional(),
@@ -115,7 +117,7 @@ export const searchWorkspaceInputSchema = workspaceSearchFiltersSchema
     nativeQueries: nativeSearchQueriesSchema
       .optional()
       .describe(
-        'Live search only: provider-native queries (Drive q, Gmail operators, Jira JQL, Confluence CQL, GitHub qualifiers, Slack RTS). Omit for simple cross-provider terms. Use the returned live guidance and account IDs.'
+        'Live search only: provider-native queries (Drive q, Gmail operators, Jira JQL, Confluence CQL, GitHub qualifiers, Slack RTS). GitHub kind commits searches commit messages with author:, committer:, author-date:, and repo: qualifiers. Omit for simple cross-provider terms. Use the returned live guidance and account IDs.'
       ),
     query: z
       .string()
@@ -138,6 +140,7 @@ export const searchWorkspaceInputSchema = workspaceSearchFiltersSchema
   .superRefine((input, context) => {
     if (
       !input.query &&
+      !input.nativeQueries?.some((query) => query.query) &&
       !input.startDate &&
       !input.endDate &&
       !input.modifiedAfter &&
@@ -146,7 +149,7 @@ export const searchWorkspaceInputSchema = workspaceSearchFiltersSchema
       context.addIssue({
         code: 'custom',
         path: ['query'],
-        message: 'Supply search terms or a date bound.',
+        message: 'Supply search terms, a native query, or a date bound.',
       })
     if (
       input.startDate &&

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

@@ -22,7 +22,9 @@ export const nativeSearchQuerySchema = z
     provider: liveSearchProviderSchema,
     query: z.string().trim().max(2000),
     accountId: z.string().min(1).max(200).optional(),
-    kind: z.enum(['issues', 'code', 'repositories', 'merge_requests', 'wiki']).optional(),
+    kind: z
+      .enum(['issues', 'code', 'repositories', 'commits', 'merge_requests', 'wiki'])
+      .optional(),
     project: z.string().min(1).max(300).optional(),
     cursor: z.string().max(4000).optional(),
     termClauses: z.array(z.string().max(500)).max(10).optional(),
@@ -127,7 +129,7 @@ export const searchWorkspaceInputSchema = workspaceSearchFiltersSchema
     nativeQueries: nativeSearchQueriesSchema
       .optional()
       .describe(
-        'Live search only: provider-native queries (Drive q, Gmail operators, Jira JQL, Confluence CQL, GitHub qualifiers, Slack RTS). Omit for simple cross-provider terms. Use the returned live guidance and account IDs.'
+        'Live search only: provider-native queries (Drive q, Gmail operators, Jira JQL, Confluence CQL, GitHub qualifiers, Slack RTS). GitHub kind commits searches commit messages with author:, committer:, author-date:, and repo: qualifiers. Omit for simple cross-provider terms. Use the returned live guidance and account IDs.'
       ),
     query: z
       .string()
@@ -150,6 +152,7 @@ export const searchWorkspaceInputSchema = workspaceSearchFiltersSchema
   .superRefine((input, context) => {
     if (
       !input.query &&
+      !input.nativeQueries?.some((query) => query.query) &&
       !input.startDate &&
       !input.endDate &&
       !input.modifiedAfter &&
@@ -158,7 +161,7 @@ export const searchWorkspaceInputSchema = workspaceSearchFiltersSchema
       context.addIssue({
         code: 'custom',
         path: ['query'],
-        message: 'Supply search terms or a date bound.',
+        message: 'Supply search terms, a native query, or a date bound.',
       })
     if (
       input.startDate &&

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

@@ -6041,7 +6041,7 @@ export const SearchWorkspace: ToolCatalogEntry = {
       },
       nativeQueries: {
         description:
-          'Live search only: provider-native queries (Drive q, Gmail operators, Jira JQL, Confluence CQL, GitHub qualifiers, Slack RTS). Omit for simple cross-provider terms. Use the returned live guidance and account IDs.',
+          'Live search only: provider-native queries (Drive q, Gmail operators, Jira JQL, Confluence CQL, GitHub qualifiers, Slack RTS). GitHub kind commits searches commit messages with author:, committer:, author-date:, and repo: qualifiers. Omit for simple cross-provider terms. Use the returned live guidance and account IDs.',
         minItems: 1,
         maxItems: 9,
         type: 'array',
@@ -6066,7 +6066,7 @@ export const SearchWorkspace: ToolCatalogEntry = {
             accountId: { type: 'string', minLength: 1, maxLength: 200 },
             kind: {
               type: 'string',
-              enum: ['issues', 'code', 'repositories', 'merge_requests', 'wiki'],
+              enum: ['issues', 'code', 'repositories', 'commits', 'merge_requests', 'wiki'],
             },
             project: { type: 'string', minLength: 1, maxLength: 300 },
             cursor: { type: 'string', maxLength: 4000 },

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

@@ -5972,7 +5972,7 @@ export const TOOL_RUNTIME_SCHEMAS: Record<string, ToolRuntimeSchemaEntry> = {
         },
         nativeQueries: {
           description:
-            'Live search only: provider-native queries (Drive q, Gmail operators, Jira JQL, Confluence CQL, GitHub qualifiers, Slack RTS). Omit for simple cross-provider terms. Use the returned live guidance and account IDs.',
+            'Live search only: provider-native queries (Drive q, Gmail operators, Jira JQL, Confluence CQL, GitHub qualifiers, Slack RTS). GitHub kind commits searches commit messages with author:, committer:, author-date:, and repo: qualifiers. Omit for simple cross-provider terms. Use the returned live guidance and account IDs.',
           minItems: 1,
           maxItems: 9,
           type: 'array',
@@ -6004,7 +6004,7 @@ export const TOOL_RUNTIME_SCHEMAS: Record<string, ToolRuntimeSchemaEntry> = {
               },
               kind: {
                 type: 'string',
-                enum: ['issues', 'code', 'repositories', 'merge_requests', 'wiki'],
+                enum: ['issues', 'code', 'repositories', 'commits', 'merge_requests', 'wiki'],
               },
               project: {
                 type: 'string',

apps/sim/lib/sim-search/live/README.md

@@ -58,7 +58,7 @@ Member mode only. Search uses the connected user's Slack real-time search grant
 
 ### GitHub
 
-Member mode searches issues, code, and repositories permitted by the connected token. Explicit repository/organization/user qualifiers narrow the user's query. Default discovery is bounded to up to 100 affiliated repositories, sent as `repo:` qualifiers in at most four batches per search kind. Code batches stay under code search's 1,000-byte query limit, issue batches are larger, and a long query that cannot fit every repository reports the searched subset. Date bounds become one `updated:start..end` range, since GitHub ORs repeated qualifiers. Provider pagination/search caps still apply. REST code search returns no file dates and accepts no date qualifier, so date-filtered searches cover issues and pull requests only.
+Member mode searches issues, code, repositories, and commits permitted by the connected token. Commits are searched only with an explicit `commits` kind; they cover each repository's default branch, and a commit read lists up to 100 changed files. Explicit repository/organization/user qualifiers narrow the user's query. Default discovery is bounded to up to 100 affiliated repositories, sent as `repo:` qualifiers in at most four batches per search kind. Code batches stay under code search's 1,000-byte query limit, issue batches are larger, and a long query that cannot fit every repository reports the searched subset. Date bounds become one `updated:start..end` range (`author-date:` for commits), since GitHub ORs repeated qualifiers. Provider pagination/search caps still apply. REST code search returns no file dates and accepts no date qualifier, so date-filtered searches cover issues and pull requests only.
 
 In service mode, an administrator connects a GitHub App installation and selects repositories one by one in Sources. Each source pins the provider-verified repository ID and may narrow code files by directory and extension. Search queries the member's own GitHub connection with `repo:` qualifiers drawn only from active sources. For each candidate, Sim checks that the current App installation still covers that repository, mints a repository-scoped read token, and compares repository and owner IDs returned under both the App and member tokens. It then checks the per-repository code filters. Reads use the member token and repeat these checks. A personal repository outside the selected sources is never searched, even if the member can access it. GitHub REST code search covers the default branch; live Sources therefore do not offer a branch setting.
 
@@ -124,7 +124,7 @@ Self-managed GitLab is resolved from the saved source's validated host/project i
 | Slack | `POST /api/assistant.search.context` | `conversations.replies` or `files.info` preview | Member only; Slack enforces the connected user's grant |
 | Jira | `POST /ex/jira/{cloudId}/rest/api/3/search/jql` | `/rest/api/3/issue/{key}` under that cloud site | Member only |
 | Confluence | `/ex/confluence/{cloudId}/wiki/rest/api/search` with CQL | `/wiki/rest/api/content/{id}` | Same site, spaces, current type/status/labels, source readability |
-| GitHub | `/search/issues`, `/search/code`, `/search/repositories` | Issue, repository, or contents endpoint for returned kind | Added repositories; installation coverage/stable IDs and code filters |
+| GitHub | `/search/issues`, `/search/code`, `/search/repositories`, `/search/commits` | Issue, repository, commit, or contents endpoint for returned kind | Added repositories; installation coverage/stable IDs and code filters |
 | GitLab | Configured `/api/v4/projects/{project}/search`, or supported date listing | Project issue/MR/wiki/file endpoint | Current request-local admin ACL evidence or saved CSV grants, plus content filters |
 | Coda | Personal MCP `search`; REST `/apis/v1/docs` title-search compatibility | MCP read allowlist; REST compatibility document/page reads | Selected parent doc and current source-token visibility; optional Enterprise org membership |
 

apps/sim/lib/sim-search/live/application.test.ts

@@ -510,6 +510,22 @@ describe('authorized live retrieval', () => {
     expect(result.retrieval.status).toBe('partial')
     expect(result.live?.accounts[0]).toMatchObject({ status: 'partial', nextCursor: 'next' })
   })
+  it('searches with only a native query and still rejects a search with neither', async () => {
+    const nativeQueries = [{ provider: 'google_drive' as const, query: "name contains 'Q4'" }]
+    await searchLiveKnowledge.execute({ principal, input: { ...input, query: '', nativeQueries } })
+    expect(mocks.search).toHaveBeenCalledExactlyOnceWith(
+      expect.anything(),
+      expect.anything(),
+      expect.objectContaining({ native: expect.objectContaining(nativeQueries[0]) })
+    )
+    await expect(
+      searchLiveKnowledge.execute({
+        principal,
+        input: { ...input, query: '', nativeQueries: [{ provider: 'google_drive', query: '' }] },
+      })
+    ).rejects.toThrow()
+    expect(mocks.search).toHaveBeenCalledOnce()
+  })
   it('rejects invalid dates before resolving provider credentials', async () => {
     await expect(
       searchLiveKnowledge.execute({

apps/sim/lib/sim-search/live/application.ts

@@ -279,8 +279,13 @@ export const searchLiveKnowledge = defineAuthorizedKnowledgeUseCase({
     const userId = requirePrincipalSubjectUserId(principal)
     if (input.organizationId) await requireOrganizationSearchAvailable(input.organizationId)
     input.signal?.throwIfAborted()
+    const queries = input.nativeQueries
+      ? nativeSearchQueriesSchema.parse(input.nativeQueries)
+      : undefined
     if (
-      (!input.query.trim() && !hasDateBounds(input.filters)) ||
+      (!input.query.trim() &&
+        !hasDateBounds(input.filters) &&
+        !queries?.some((query) => query.query)) ||
       input.query.length > 2000 ||
       !Number.isInteger(input.topK) ||
       input.topK < 1 ||
@@ -296,9 +301,6 @@ export const searchLiveKnowledge = defineAuthorizedKnowledgeUseCase({
       Date.parse(filters.startDate) >= Date.parse(filters.endDate)
     )
       throw new OrchestrationError('validation', 'endDate must be after startDate')
-    const queries = input.nativeQueries
-      ? nativeSearchQueriesSchema.parse(input.nativeQueries)
-      : undefined
     if (queries?.some((query) => !query.query) && !hasDateBounds(filters))
       throw new OrchestrationError('validation', 'Empty native queries require a date bound')
     const searchSignal = input.signal

apps/sim/lib/sim-search/live/github.ts

@@ -13,7 +13,28 @@ import type {
   NativeSearchInput,
 } from '@/lib/sim-search/live/types'
 
+/** A commit as a document: its message is the content and its author date the timeline. */
+function commitDocument(
+  row: Record<string, unknown>,
+  container = string(object(row.repository).full_name)
+): NativeDocument {
+  const commit = object(row.commit)
+  const message = string(commit.message)
+  const authored = Date.parse(string(object(commit.author).date))
+  return {
+    id: string(row.sha),
+    container,
+    kind: 'commits',
+    title: `${container} · ${message.split('\n')[0] || string(row.sha).slice(0, 7)}`,
+    url: string(row.html_url),
+    content: message,
+    ...(Number.isFinite(authored) ? { modifiedAt: new Date(authored).toISOString() } : {}),
+    author: string(object(row.author).login) || string(object(commit.author).name),
+  }
+}
+
 function githubDocument(row: Record<string, unknown>, kind: string): NativeDocument {
+  if (kind === 'commits') return commitDocument(row)
   const repository = object(row.repository)
   const repositoryUrl = string(row.repository_url)
   const container =
@@ -59,17 +80,31 @@ const GITHUB_ISSUE_QUERY_BYTES = 3800
 /** Batches per kind; each code batch spends one of GitHub's ten code searches per minute. */
 const GITHUB_MAX_REPOSITORY_BATCHES = 4
 
-type GitHubKind = 'issues' | 'code' | 'repositories'
+/** Changed files listed when reading a commit, GitHub's default page size. */
+const GITHUB_COMMIT_FILES = 30
+
+const GITHUB_KINDS = ['issues', 'code', 'repositories', 'commits'] as const
+type GitHubKind = (typeof GITHUB_KINDS)[number]
+const isGitHubKind = (kind: string | undefined): kind is GitHubKind =>
+  GITHUB_KINDS.some((supported) => supported === kind)
+/**
+ * The date each kind filters and sorts by: issues by last update, commits by author date. Code
+ * and repository search take no date qualifier.
+ */
+const GITHUB_DATE_FIELD: Partial<Record<GitHubKind, 'updated' | 'author-date'>> = {
+  issues: 'updated',
+  commits: 'author-date',
+}
 const CODE_EXCLUDED_BY_DATES = 'Code has no dates and is excluded from date-filtered searches.'
 
 /** Code has no file dates, so a date-bounded search covers issues and pull requests only. */
 function githubKinds(input: NativeSearchInput): GitHubKind[] {
   const kind = input.native?.kind
-  if (kind === 'issues' || kind === 'code' || kind === 'repositories') return [kind]
+  if (isGitHubKind(kind)) return [kind]
   if (kind)
     throw new NativeSearchError(
       'unavailable',
-      'GitHub search supports issues, code, or repositories.'
+      `GitHub search supports these kinds: ${GITHUB_KINDS.join(', ')}.`
     )
   return hasDateBounds(input.filters) ? ['issues'] : ['issues', 'code']
 }
@@ -240,17 +275,17 @@ export async function searchGitHub(
     )
   const dates = nativeDateBounds(input)
   const text = nativeText(input)
-  /** GitHub ORs repeated qualifiers, so both bounds must share one `updated:` range. */
-  const updated =
+  /** GitHub ORs repeated qualifiers, so both bounds share one range qualifier. */
+  const dateField = GITHUB_DATE_FIELD[kind]
+  const dateRange =
     dates.start && dates.end
-      ? `updated:${dates.start}..${dates.end}`
+      ? `${dateField}:${dates.start}..${dates.end}`
       : dates.start
-        ? `updated:>=${dates.start}`
+        ? `${dateField}:>=${dates.start}`
         : dates.end
-          ? `updated:<=${dates.end}`
+          ? `${dateField}:<=${dates.end}`
           : ''
-  const datedQuery =
-    kind === 'issues' ? [groupGitHubText(text), updated].filter(Boolean).join(' ') : text
+  const datedQuery = dateField ? [groupGitHubText(text), dateRange].filter(Boolean).join(' ') : text
   if (githubTextLength(text) > GITHUB_TEXT_CHARACTERS)
     throw new NativeSearchError(
       'unavailable',
@@ -266,8 +301,8 @@ export async function searchGitHub(
         q: hasDateBounds(input.filters) ? datedQuery : text,
         per_page: String(input.limit),
         page,
-        ...(kind === 'issues' && dateSortDirection(input.filters)
-          ? { sort: 'updated', order: dateSortDirection(input.filters) }
+        ...(dateField && dateSortDirection(input.filters)
+          ? { sort: dateField, order: dateSortDirection(input.filters) }
           : {}),
       },
     })
@@ -333,6 +368,33 @@ export async function readGitHub(
     const row = object(await client.json(path))
     return githubDocument(row, 'repositories')
   }
+  if (kind === 'commits') {
+    if (!/^[0-9a-f]{7,40}$/i.test(id))
+      throw new NativeSearchError('unavailable', 'Invalid GitHub commit reference.')
+    /** Each changed file carries its patch, so the page bounds the response size. */
+    const row = object(
+      await client.json(`${path}/commits/${id}`, {
+        query: { per_page: String(GITHUB_COMMIT_FILES) },
+      })
+    )
+    const files = array(row.files).map(
+      (file) =>
+        `${string(file.status)} ${string(file.filename)} (+${string(file.additions)} -${string(file.deletions)})`
+    )
+    const document = commitDocument(row, repository)
+    return {
+      ...document,
+      content: [
+        document.content,
+        files.length ? `Files changed:\n${files.join('\n')}` : '',
+        files.length === GITHUB_COMMIT_FILES
+          ? `At most ${GITHUB_COMMIT_FILES} changed files are listed.`
+          : '',
+      ]
+        .filter(Boolean)
+        .join('\n\n'),
+    }
+  }
   if (!/^\d+$/.test(id))
     throw new NativeSearchError('unavailable', 'Invalid GitHub issue reference.')
   return githubDocument(object(await client.json(`${path}/issues/${id}`)), 'issues')