fix(search): search nothing for a Confluence query with no words and stop the Assistant retrying unsearchable searches

A query left with no words after the colon and backslash become spaces would
list every Confluence page, so it now searches nothing on both the member and
service-account paths. The Assistant search tool reports a search no account
served as nothing searched instead of advising a narrower retry, and the
organization search E2E asserts the connect prompt the unconnected member sees.

Author: waleedlatif1

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

@@ -246,6 +246,19 @@ describe('Assistant retrieval tools', () => {
     )
     expect(JSON.stringify(result)).not.toContain(secret)
   })
+  it('tells the model a search no account served searched nothing, not to narrow and retry', async () => {
+    mocks.search.mockResolvedValueOnce({
+      query: 'roadmap',
+      retrieval: { status: 'partial', timedOutLegs: [] },
+      results: [],
+      live: { backend: 'live', accounts: [], guidance: 'Nothing was searched: connect one.' },
+    })
+    const result = await searchWorkspaceServerTool.execute({ query: 'roadmap' }, context)
+    expect(result).toMatchObject({
+      success: true,
+      message: expect.stringMatching(/^Nothing was searched: (?!.*more specific query)/),
+    })
+  })
   it('rejects untrusted contexts, incompatible sources and out-of-scope document reads', async () => {
     expect(
       await searchWorkspaceServerTool.execute(

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

@@ -85,9 +85,13 @@ export const searchWorkspaceServerTool: BaseServerTool = {
                   ...liveInput,
                   workspaceId: scope.workspaceId,
                 })
+          /** Every searched account reports a status, so none means no account could serve it. */
+          const nothingSearched = data.live?.accounts.length === 0
           return {
             success: true,
-            message: `${data.retrieval.status === 'partial' ? 'Search coverage is incomplete. Continue with a more specific query or source filter; these results cannot establish absence or completeness. ' : ''}Found ${data.results.length} live results. Read a documentId when its passage does not answer the question or more context is needed. ${CITATION_INSTRUCTION}`,
+            message: nothingSearched
+              ? 'Nothing was searched: no connected account can serve this search, so it says nothing about what exists. Follow the first line of live.guidance (connect an account, or drop the source filter) instead of retrying the same search.'
+              : `${data.retrieval.status === 'partial' ? 'Search coverage is incomplete. Continue with a more specific query or source filter; these results cannot establish absence or completeness. ' : ''}Found ${data.results.length} live results. Read a documentId when its passage does not answer the question or more context is needed. ${CITATION_INSTRUCTION}`,
             data: {
               ...data,
               results: data.results.map((item) => ({

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

@@ -244,6 +244,13 @@ describe('Jira any-term search', () => {
     const page = await searchAtlassian(api, 'confluence', { ...query, textMatch: 'any' })
     expect(page.documents.map((document) => document.id)).toEqual(['5'])
   })
+
+  it('searches nothing for a Confluence query with no word left, which would list every page', async () => {
+    const api = client({})
+    const page = await searchAtlassian(api, 'confluence', { ...query, query: String.raw` :\ ` })
+    expect(page.documents).toEqual([])
+    expect(api.json).not.toHaveBeenCalled()
+  })
 })
 
 describe('Confluence plain search ranking', () => {

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

@@ -132,6 +132,12 @@ export async function searchAtlassian(
   provider: 'jira' | 'confluence',
   input: NativeSearchInput
 ): Promise<NativePage> {
+  const text = nativeText(input)
+  const confluenceClause =
+    provider === 'confluence' && text && !input.native?.query
+      ? confluenceTextClause(text, input.textMatch)
+      : ''
+  if (confluenceClause === undefined) return { documents: [] }
   const allSites = (await sites(client)).filter(
     (site) =>
       !input.policy?.sites.length ||
@@ -166,7 +172,6 @@ export async function searchAtlassian(
       .join(' AND ') || undefined
   const direction = dateSortDirection(input.filters)
   const order = direction ? `${field} ${direction.toUpperCase()}` : undefined
-  const text = nativeText(input)
   // Jira's any-term JQL matches issues sharing two common words; like minimum-should-match,
   // keep only issues holding at least half the terms.
   const relevant = input.textMatch === 'any' && text ? holdsHalfTheTerms(text) : undefined
@@ -234,13 +239,15 @@ export async function searchAtlassian(
         }
       }
       const exact = await searchPages(
-        text ? confluenceTextClause(text, input.textMatch) : '',
+        confluenceClause,
         input.native?.cursor && single ? input.native.cursor : undefined
       )
+      const relevance = confluenceTextClause(text, 'any')
       // An all-terms match larger than a page comes back by raw term matches; rank it by relevance.
-      if (!text || input.native || order || input.textMatch === 'any' || !exact.next) return exact
+      if (!relevance || input.native || order || input.textMatch === 'any' || !exact.next)
+        return exact
       try {
-        const ranked = await searchPages(confluenceTextClause(text, 'any'))
+        const ranked = await searchPages(relevance)
         return { ...ranked, next: undefined, ranking: 'relevance' }
       } catch {
         // A cancelled search stops here; any other failure keeps the all-terms page.

apps/sim/lib/sim-search/live/service-session.ts

@@ -146,6 +146,7 @@ export async function createLiveServiceSession(input: {
         const query =
           search.native?.query ??
           (search.query ? confluenceTextClause(search.query, search.textMatch) : '')
+        if (query === undefined) return null
         return {
           ...search,
           native: { provider, ...search.native, query: scopeAtlassianQuery(query, scope) },

apps/sim/lib/sim-search/live/text-match.test.ts

@@ -66,7 +66,7 @@ describe('text match clauses', () => {
     expect(confluenceTextClause(String.raw`C:\builds\out`, 'any')).toBe(
       'siteSearch ~ "C  builds out"'
     )
-    expect(confluenceTextClause(String.raw` :\ `, 'all')).toBe('')
+    expect(confluenceTextClause(String.raw` :\ `, 'all')).toBeUndefined()
   })
 
   it('falls back to the all-terms clause when fewer than two terms are distinctive', () => {

apps/sim/lib/sim-search/live/text-match.ts

@@ -151,13 +151,15 @@ export function jiraTextClause(text: string, match: TextMatch = 'all'): string {
  * relevance search, which ranks a page with the query's rarer words above the many pages sharing
  * one common word, where a `text ~` OR query ranks them the other way round.
  *
- * Confluence answers HTTP 400 to a text query holding a colon or backslash, even escaped, and its
- * analyzer splits words on both, so they go as spaces. Nothing left to match means no clause.
+ * Confluence answers HTTP 400 to the escaped colon or backslash this phrase would carry, and its
+ * analyzer splits words on both, so they go as spaces. `undefined` means nothing is left to match:
+ * Confluence would list every page for such a phrase, so callers search nothing instead.
  */
-export function confluenceTextClause(text: string, match: TextMatch = 'all'): string {
+export function confluenceTextClause(text: string, match: TextMatch = 'all'): string | undefined {
   const searchable = text.replace(/[:\\]/g, ' ').trim()
-  if (!searchable) return ''
-  return `${match === 'any' ? 'siteSearch' : 'text'} ~ ${escapeSearchPhrase(searchable)}`
+  return searchable
+    ? `${match === 'any' ? 'siteSearch' : 'text'} ~ ${escapeSearchPhrase(searchable)}`
+    : undefined
 }
 
 /** How many distinct terms occur as whole words in the text. */

apps/sim/scripts/test-organization-search-e2e.ts

@@ -445,7 +445,15 @@ try {
       const body = toRecord(response.request().postDataJSON())
       assert.equal(body.organizationId, organizationId)
       assert.equal(body.query, 'Launch notes')
-      await expect(page.getByText('Search found no results.', { exact: true })).toBeVisible()
+      // The seeded member has connected no account, so nothing was searched.
+      await expect(
+        page.getByText('Connect your apps to search them.', { exact: true })
+      ).toBeVisible()
+      await expect(page.getByText('Search found no results.', { exact: true })).toHaveCount(0)
+      await expect(page.getByRole('link', { name: 'Connect apps' })).toHaveAttribute(
+        'href',
+        `${root}/integrations`
+      )
       await expect(page).toHaveURL(
         (url) => url.pathname === `${root}/search` && url.searchParams.get('q') === 'Launch notes'
       )