zames_pro 2.70.0 → 2.72.0

package/scripts/{render-demo.mjs → render-demo.mts}

@@ -7,7 +7,7 @@
 // theme palette (src/theme.ts) with the DejaVu Sans Mono font that the
 // container actually ships, then screenshots it with headless Chromium.
 //
-// Run:  node scripts/render-demo.mjs
+// Run:  npm run render-demo  (tsx scripts/render-demo.mts)
 //
 // Isolation: chromium.launch() uses a THROWAWAY profile, never ~/.zames/profile
 // (see AGENTS.md — never touch the live agent's browser/profile).
@@ -33,9 +33,9 @@ const C = {
   toggleOn: '#6fd0b0',
 }
 
-const esc = (s) =>
+const esc = (s: unknown): string =>
   String(s).replace(/&/g, '&amp;').replace(/</g, '&lt;').replace(/>/g, '&gt;')
-const span = (color, text, bold = false) =>
+const span = (color: string, text: string, bold = false): string =>
   '<span style="color:' +
   color +
   (bold ? ';font-weight:700' : '') +
@@ -44,13 +44,18 @@ const span = (color, text, bold = false) =>
   '</span>'
 
 // One rendered session, line by line (mirrors the real banner + answer flow).
-const lines = []
-const push = (html) => lines.push('<div class="l">' + html + '</div>')
+const lines: string[] = []
+const push = (html: string): void => {
+  lines.push('<div class="l">' + html + '</div>')
+}
 const blank = () => push('&nbsp;')
 
 push(span(C.system, 'Working directory: ') + span(C.dir, '/home/dev/zames_pro'))
 push(span(C.system, 'Version: ') + span(C.user, '2.64.1'))
-push(span(C.system, 'Transcript: ') + span(C.dim, '~/.zames/logs/zames_pro-2026-10-07.log'))
+push(
+  span(C.system, 'Transcript: ') +
+    span(C.dim, '~/.zames/logs/zames_pro-2026-10-07.log'),
+)
 blank()
 push(
   span(
@@ -78,11 +83,7 @@ push(
     span(C.user, 'Show the version from package.json.'),
 )
 blank()
-push(
-  span(C.tool, '🔧 Read') +
-    ' ' +
-    span(C.dim, '{"path":"package.json"}'),
-)
+push(span(C.tool, '🔧 Read') + ' ' + span(C.dim, '{"path":"package.json"}'))
 push(
   span(C.toolResult, '   → ') +
     span(

package/scripts/{self-smoke.mjs → self-smoke.mts}

@@ -42,13 +42,19 @@ fs.mkdirSync(WORK, { recursive: true })
 
 // --- project modules (dynamic: they read HOME at import) --------------------
 const { DeepSeekBrowser } = await import('../src/browser.ts')
-const { createTools } = await import('../src/tools.ts')
+const { createTools, filterToolsForReadOnly } = await import('../src/tools.ts')
 const { runAgentLoop } = await import('../src/agent-loop.ts')
+const { createSubagentRunner } = await import('../src/subagent.ts')
 const { loadConfig } = await import('../src/config.ts')
 const { Transcript } = await import('../src/transcript.ts')
 
-const results = []
-const record = (name, ok, note) => {
+interface SmokeResult {
+  name: string
+  ok: boolean
+  note?: string
+}
+const results: SmokeResult[] = []
+const record = (name: string, ok: boolean, note?: string): void => {
   results.push({ name, ok, note })
   console.log((ok ? 'PASS ' : 'FAIL ') + name + (note ? '  — ' + note : ''))
 }
@@ -148,14 +154,89 @@ async function main() {
     askDeadlineMs: cfg.browser.askDeadlineMs,
     maxAfterToolRetries: cfg.browser.maxAfterToolRetries,
   })
-  record('follow-up answered', outC.trim().length > 0, JSON.stringify(outC.slice(0, 80)))
+  record(
+    'follow-up answered',
+    outC.trim().length > 0,
+    JSON.stringify(outC.slice(0, 80)),
+  )
+  const onSubagent = createSubagentRunner({
+    browser,
+    workdir: WORK,
+    locale: 'ru',
+    runAgentLoop,
+    buildTools: (readOnly) => {
+      const base = createTools(WORK, {})
+      return readOnly ? filterToolsForReadOnly(base) : base
+    },
+    transcript,
+    maxSubagents: 2,
+    askDeadlineMs: cfg.browser.askDeadlineMs,
+    maxAfterToolRetries: cfg.browser.maxAfterToolRetries,
+  })
+
+  // Scenario D — subagents (N36): the model delegates via Task; the runner
+  // opens a separate chat, returns only the report, and restores the parent
+  // chat. Requires browser.subagents (off by default).
+  const subTools = createTools(WORK, { subagents: true })
+  const outD = await runAgentLoop({
+    browser,
+    tools: subTools,
+    task:
+      'Use the Task tool with subagent_type "explore" to find which word is written in the file smoke.txt. ' +
+      'prompt the subagent in a self-contained way. Then, based on its report, call respond with that word.',
+    workdir: WORK,
+    maxIterations: 20,
+    freshChat: true,
+    sendSystemPrompt: true,
+    transcript,
+    onSubagent,
+    onToolCall: (n) => console.log('  tool: ' + n),
+    onAssistantMessage: (m) => console.log('  answer: ' + m),
+    onWarning: (m) => console.log('  warn: ' + m),
+    locale: 'ru',
+    askDeadlineMs: cfg.browser.askDeadlineMs,
+    maxAfterToolRetries: cfg.browser.maxAfterToolRetries,
+  })
+  record(
+    'subagent: parent used the report',
+    /HELLO_SMOKE/i.test(outD),
+    JSON.stringify(outD.slice(0, 120)),
+  )
+
+  // Scenario E — background Bash + BashOutput (N33): start a process that
+  // keeps running, poll its output, then kill it.
+  const outE = await runAgentLoop({
+    browser,
+    tools,
+    task:
+      'Do exactly this, then report: (1) Start the Bash command ' +
+      '`echo BG_MARKER_7 && sleep 120` in the BACKGROUND (run_in_background=true). ' +
+      '(2) Poll it once with BashOutput. (3) Kill it with BashOutput kill=true. ' +
+      'Then call respond with one short line containing BG_MARKER_7.',
+    workdir: WORK,
+    maxIterations: 20,
+    freshChat: true,
+    sendSystemPrompt: true,
+    transcript,
+    onToolCall: (n) => console.log('  tool: ' + n),
+    onAssistantMessage: (m) => console.log('  answer: ' + m),
+    onWarning: (m) => console.log('  warn: ' + m),
+    locale: 'ru',
+    askDeadlineMs: cfg.browser.askDeadlineMs,
+    maxAfterToolRetries: cfg.browser.maxAfterToolRetries,
+  })
+  record(
+    'background bash: started, polled and killed',
+    /BG_MARKER_7/.test(outE),
+    JSON.stringify(outE.slice(0, 120)),
+  )
 }
 
 let failed = false
 try {
   await main()
 } catch (e) {
-  record('run', false, e.message)
+  record('run', false, (e as Error).message)
   failed = true
 } finally {
   await browser.close().catch(() => {})