improvement(agent-cli): skip non-text files in grep instead of marking the search incomplete

apps/sim/lib/mothership/agent-cli/engines/universal-grep.test.ts

@@ -524,6 +524,121 @@ describe('search authority and completeness', () => {
     expect(result.stderr).toContain('broken')
   })
 
+  it('skips files with no text extractor without fetching them or failing the search', async () => {
+    const requested: string[] = []
+    const result = await runEngine(
+      'grep',
+      ['needle'],
+      runtimeWith(
+        {
+          '/api/v2/files': {
+            data: [
+              { id: 'notes', name: 'notes.md', type: 'text/markdown' },
+              { id: 'report', name: 'report.pdf', type: 'application/pdf' },
+              { id: 'font', name: 'brand.woff2', type: 'font/woff2' },
+              { id: 'clip', name: 'demo.mp4', type: 'video/mp4' },
+            ],
+            nextCursor: null,
+          },
+          '/api/v2/files/notes/text': { data: { text: 'needle', degraded: false } },
+          '/api/v2/files/report/text': { data: { text: 'a needle', degraded: false } },
+        },
+        requested
+      ),
+      { scope: 'files' }
+    )
+    expect(requested.filter((path) => path !== '/api/v2/files').sort()).toEqual([
+      '/api/v2/files/notes/text',
+      '/api/v2/files/report/text',
+    ])
+    expect(result.exitCode).toBe(0)
+    expect(result.stdout).toBe(
+      'files/notes.md (notes):1: needle\nfiles/report.pdf (report):1: a needle'
+    )
+    expect(result.stderr).toBe(
+      'Skipped 2 files with no searchable text (images, media, archives, and other binary formats).'
+    )
+  })
+
+  it('keeps real read failures incomplete beside skipped non-text files', async () => {
+    const result = await runEngine(
+      'grep',
+      ['needle'],
+      runtimeWith({
+        '/api/v2/files': {
+          data: [
+            { id: 'broken', name: 'broken.txt', type: 'text/plain' },
+            { id: 'logo', name: 'logo.png', type: 'image/png' },
+          ],
+          nextCursor: null,
+        },
+      }),
+      { scope: 'files' }
+    )
+    expect(result.exitCode).toBe(1)
+    expect(result.stderr).toMatch(/^Search incomplete \(1 issue\)/)
+    expect(result.stderr).toContain('files/broken.txt (broken): could not read text')
+    expect(result.stderr).toContain('Skipped 1 file with no searchable text')
+  })
+
+  it('names a selected non-text file as searched-and-skipped, not an unknown selector', async () => {
+    const requested: string[] = []
+    const result = await runEngine(
+      'grep',
+      ['needle'],
+      runtimeWith(
+        {
+          '/api/v2/files': {
+            data: [{ id: 'logo', name: 'logo.png', type: 'image/png' }],
+            nextCursor: null,
+          },
+        },
+        requested
+      ),
+      { scope: 'files', in: 'files/logo' }
+    )
+    expect(requested).toEqual(['/api/v2/files'])
+    expect(result.exitCode).toBe(0)
+    expect(result.stdout).toContain('No matches')
+    expect(result.stderr).toContain('Skipped 1 file with no searchable text')
+  })
+
+  it('spends the file-read limit only on files with text', async () => {
+    const files = [
+      ...Array.from({ length: 300 }, (_, i) => ({
+        id: `img${i}`,
+        name: `img${i}.png`,
+        type: 'image/png',
+      })),
+      { id: 'notes', name: 'notes.txt', type: 'text/plain' },
+    ]
+    const result = await runEngine(
+      'grep',
+      ['needle'],
+      {
+        ...runtimeWith({}),
+        client: {
+          request: async <T>(
+            path: string,
+            options?: { query?: Record<string, string> }
+          ): Promise<T> => {
+            if (path === '/api/v2/files') {
+              const offset = Number(options?.query?.cursor ?? 0)
+              return {
+                data: files.slice(offset, offset + 100),
+                nextCursor: offset + 100 < files.length ? String(offset + 100) : null,
+              } as T
+            }
+            if (path === '/api/v2/files/notes/text') return { data: { text: 'needle' } } as T
+            throw new Error(`Unexpected request: ${path}`)
+          },
+        },
+      },
+      { scope: 'files', count: true }
+    )
+    expect(result).toMatchObject({ exitCode: 0, stdout: '1 (files=1)' })
+  })
+
   it('reports truncated extraction even when a returned prefix has a match', async () => {
     const result = await runEngine(
       'grep',

apps/sim/lib/mothership/agent-cli/engines/universal-grep.ts

@@ -11,9 +11,9 @@ import {
   type AgentCliFlags,
   type AgentCliRuntime,
   agentCliFail,
-  agentCliOk,
 } from '@/lib/mothership/agent-cli/types'
 import { buildIntegrationToolSchemas } from '@/lib/mothership/chat/payload'
+import { workspaceFileTextFormat } from '@/lib/workspace-files/text-format'
 import { projectResolvedSecretModelContent } from '@/executor/utils/resolved-secret-content-projection'
 import { OPAQUE_RESOLVED_SECRET_REPLACEMENT } from '@/executor/utils/resolved-secret-matcher'
 import type { ResolvedSecretTraceRegistry } from '@/executor/utils/resolved-secret-trace-registry'
@@ -61,6 +61,7 @@ const PLATFORM_SCOPES: ReadonlySet<Scope> = new Set<Scope>(['blocks', 'tools'])
 
 interface GrepRuntime extends AgentCliRuntime {
   reportIncomplete(message: string): void
+  reportNonText(): void
 }
 
 /**
@@ -370,22 +371,44 @@ const FETCHERS: Record<
   credentials: async (_runtime, item) => render('credentials', item.id, item.label, item.raw),
 }
 
-function concurrencyFor(scope: Scope): number {
-  return scope === 'files' ? FILE_READ_CONCURRENCY : FETCH_CONCURRENCY
+/**
+ * A listed file the agent's text read has no extractor for (fonts, media, archives, images), by
+ * the same gate that read applies. A listing without a name cannot say, so that file is read.
+ */
+function lacksText(file: unknown): boolean {
+  if (!isRecordLike(file)) return false
+  const name = str(file.name)
+  return name !== undefined && !workspaceFileTextFormat({ name, type: str(file.type) ?? '' })
 }
 
 async function fetchAll(
   runtime: GrepRuntime,
   scope: Scope,
   entries: IndexEntry[]
 ): Promise<GrepMaterialized[]> {
-  if (scope === 'files' && entries.length > MAX_GREP_FILES) {
+  if (scope !== 'files') {
+    return mapConcurrent(entries, FETCH_CONCURRENCY, (item) => FETCHERS[scope](runtime, item))
+  }
+  // Skipped files keep their identity, so `--in` naming one is not an unknown selector.
+  const skipped: GrepMaterialized[] = []
+  const readable: IndexEntry[] = []
+  for (const item of entries) {
+    if (lacksText(item.raw)) {
+      runtime.reportNonText()
+      skipped.push({ scope, id: item.id, label: item.label, text: null })
+    } else readable.push(item)
+  }
+  if (readable.length > MAX_GREP_FILES) {
     runtime.reportIncomplete(
-      `files: searched the first ${MAX_GREP_FILES} of ${entries.length} selected files; narrow with --in files/<name-or-id>.`
+      `files: searched the first ${MAX_GREP_FILES} of ${readable.length} selected files; narrow with --in files/<name-or-id>.`
     )
   }
-  const selected = scope === 'files' ? entries.slice(0, MAX_GREP_FILES) : entries
-  return mapConcurrent(selected, concurrencyFor(scope), (item) => FETCHERS[scope](runtime, item))
+  const read = await mapConcurrent(
+    readable.slice(0, MAX_GREP_FILES),
+    FILE_READ_CONCURRENCY,
+    (item) => FETCHERS.files(runtime, item)
+  )
+  return [...read, ...skipped]
 }
 
 /** A whole world, for a search with no `--in`: every resource the index lists. */
@@ -566,10 +589,15 @@ export interface GrepSource {
   registry?: AgentCliRuntime['resolvedSecretTraceRegistry']
   /** The only scopes this corpus holds, and the refusal for a search that names another. */
   scopeLimit?: { scopes: readonly Scope[]; refusal: string }
+  /**
+   * `reportNonText` counts a file skipped unread because it has no text to extract: a note,
+   * not an incomplete search. The skipped file is still returned with null text.
+   */
   materialize(
     scopes: Scope[],
     nameFilter: string | undefined,
-    reportIncomplete: (message: string) => void
+    reportIncomplete: (message: string) => void,
+    reportNonText: () => void
   ): Promise<GrepMaterialized[]>
 }
 
@@ -624,15 +652,24 @@ export async function executeGrep(positionals: string[], flags: AgentCliFlags, s
     issueCount++
     if (issues.length < 10) issues.push(clip(message))
   }
-  const finish = (stdout: string) =>
-    issueCount === 0
-      ? agentCliOk(stdout)
-      : {
-          exitCode: 1,
-          stdout,
-          stderr: `Search incomplete (${issueCount} issue${issueCount === 1 ? '' : 's'}); matches and counts cover only the content read.\n${issues.join('\n')}${issueCount > issues.length ? `\n${issueCount - issues.length} further issues omitted.` : ''}`,
-        }
-  const candidates = await source.materialize(searched, nameFilter, reportIncomplete)
+  let nonTextCount = 0
+  const reportNonText = () => {
+    nonTextCount++
+  }
+  const finish = (stdout: string) => {
+    const skipped = nonTextCount
+      ? `Skipped ${nonTextCount} file${nonTextCount === 1 ? '' : 's'} with no searchable text (images, media, archives, and other binary formats).`
+      : ''
+    const incomplete =
+      issueCount &&
+      `Search incomplete (${issueCount} issue${issueCount === 1 ? '' : 's'}); matches and counts cover only the content read.\n${issues.join('\n')}${issueCount > issues.length ? `\n${issueCount - issues.length} further issues omitted.` : ''}`
+    return {
+      exitCode: issueCount ? 1 : 0,
+      stdout,
+      stderr: [incomplete, skipped].filter(Boolean).join('\n'),
+    }
+  }
+  const candidates = await source.materialize(searched, nameFilter, reportIncomplete, reportNonText)
   source.signal?.throwIfAborted()
   /**
    * A resource nothing in the searched worlds answers to is a wrong selector, not a
@@ -721,8 +758,8 @@ export const universalGrepCommand: AgentCliEngine = {
     executeGrep(positionals, flags, {
       signal: runtime.signal,
       registry: runtime.resolvedSecretTraceRegistry,
-      async materialize(scopes, nameFilter, reportIncomplete) {
-        const bounded = gatedRuntime({ ...runtime, reportIncomplete })
+      async materialize(scopes, nameFilter, reportIncomplete, reportNonText) {
+        const bounded = gatedRuntime({ ...runtime, reportIncomplete, reportNonText })
         return nameFilter
           ? materializeWithin(bounded, scopes, nameFilter)
           : (await settleAll(scopes.map((scope) => materializeScope(bounded, scope)))).flat()

apps/sim/lib/mothership/agent-cli/project-file-grep.test.ts

@@ -105,6 +105,23 @@ describe('Project grep corpus', () => {
     expect(list).toHaveBeenCalledTimes(1)
     expect(result.stderr).toContain(`searched the first ${MAX_GREP_FILES} selected files`)
   })
+  it('skips a file with no text extractor without reading it or failing the search', async () => {
+    list.mockResolvedValue({
+      files: [
+        { id: 'file', name: 'notes.txt', type: 'text/plain', folderPath: '/' },
+        { id: 'clip', name: 'notes-demo.mp4', type: 'video/mp4', folderPath: '/' },
+      ],
+      nextKeys: null,
+    })
+    const result = await grepProjectFiles(invocation, context(), 'project')
+    expect(textRead).toHaveBeenCalledTimes(1)
+    expect(result).toEqual({
+      exitCode: 0,
+      stdout: 'files/notes.txt (file):2: needle',
+      stderr:
+        'Skipped 1 file with no searchable text (images, media, archives, and other binary formats).',
+    })
+  })
   it('marks truncated inventory incomplete instead of claiming a missing selector has no matches', async () => {
     list.mockResolvedValue({ files: [], nextKeys: [{ value: 'cursor', direction: 'asc' }] })
     const result = await grepProjectFiles(invocation, context(), 'project')