Further Harness Updates

packages/bare-os-seeder/kernel/bin/agent

@@ -1601,6 +1601,78 @@ function bareAgentLooksLikeGuestCliRequest(text) {
   )
 }
 
+/**
+ * User asked for a mutating/action job, not a definition.
+ * @param {string} text
+ */
+function bareAgentLooksLikeActionRequest(text) {
+  const s = String(text || '').toLowerCase()
+  if (!s.trim()) return false
+  return (
+    /\b(create|write|make|add|put|place|move|rename|copy|delete|remove|fix|implement|build|install|edit|patch|run)\b/.test(
+      s
+    ) ||
+    /\b(director(y|ies)|files?|scripts?|bin utils?)\b/.test(s)
+  )
+}
+
+/**
+ * Assistant is asking the operator to choose instead of finishing.
+ * @param {string} text
+ */
+function bareAgentLooksLikeQuestionDump(text) {
+  const s = String(text || '')
+  if (!s.trim()) return false
+  const low = s.toLowerCase()
+  if (
+    /would you like( me)? to/.test(low) ||
+    /do you want me to/.test(low) ||
+    /should i (create|move|rename|run|write|add|make)/.test(low) ||
+    /you('d| would) need to:/.test(low) ||
+    /the user might need to/.test(low) ||
+    /let me know if you want/.test(low) ||
+    /shall i /.test(low)
+  ) {
+    return true
+  }
+  if (
+    /would you like/.test(low) &&
+    /\n\s*1[\).:]/.test(s) &&
+    /\n\s*2[\).:]/.test(s)
+  ) {
+    return true
+  }
+  return false
+}
+
+/**
+ * Claimed a file exists / was named without a verifying tool result.
+ * @param {string} text
+ */
+function bareAgentLooksLikeUnverifiedFileClaim(text) {
+  const s = String(text || '').toLowerCase()
+  if (!s.trim()) return false
+  return (
+    /has been (created|placed|moved|renamed|written|copied) (successfully )?(in|to|at|into)/.test(
+      s
+    ) ||
+    /the file will be named/.test(s) ||
+    /is now in the (correct |right )?location/.test(s) ||
+    /scripts? (are|is) now in/.test(s)
+  )
+}
+
+/**
+ * Permission-style ask_user_question payload (create/move/rename).
+ * @param {unknown} questions
+ */
+function bareAgentLooksLikePermissionQuestions(questions) {
+  const blob = JSON.stringify(questions || '').toLowerCase()
+  return /would you like|do you want me to|should i |create a script|move the|rename the|add additional/.test(
+    blob
+  )
+}
+
 /** ~/.agent paths, config, history trim, man digest (preamble for /bin/agent). */
 
 /**
@@ -7121,11 +7193,16 @@ function bareAgentToolDefinitions() {
       function: {
         name: 'run_js_script',
         description:
-          'Run custom guest JavaScript only. Do not use this for curl, wget, ip, ping, or any shell — call run_command. Do not use this to replace read_file, write_file, list_directory, or run_command. Node is not installed. Writes ~/.agent/_tmp_agent_run.mjs and runs it on the Bare kernel. ctx and argv are already injected — never declare them, never export. Body must be: async function run(ctx, argv) { const vfs = ctx.vfs; ... }',
+          'Run custom guest JavaScript only. Do not use this for curl, wget, ip, ping, or any shell — call run_command. Do not use this to replace read_file, write_file, list_directory, or run_command. Without path=, this ONLY writes a throwaway ~/.agent/_tmp_agent_run.mjs — that is not a user file. Pass path=~/test/name.js (or call write_file) to persist. Node is not installed. ctx and argv are already injected — never declare them, never export. Body must be: async function run(ctx, argv) { const vfs = ctx.vfs; ... }',
         parameters: {
           type: 'object',
           properties: {
-            code: { type: 'string', description: 'Full ESM/CommonJS script body' }
+            code: { type: 'string', description: 'Full ESM/CommonJS script body' },
+            path: {
+              type: 'string',
+              description:
+                'If set, persist the script here (absolute or ~/...) then run it. Use this when the user asked you to create a file.'
+            }
           },
           required: ['code']
         }
@@ -8663,7 +8740,8 @@ async function bareAgentDispatchTool(o) {
       'destination',
       'target',
       'old_path',
-      'new_path'
+      'new_path',
+      'save_as'
     ]
     let cwdForExpand = resolvedHome
     if (paths && paths.lastCwd && typeof bareAgentReadTextFile === 'function') {
@@ -9197,6 +9275,18 @@ async function bareAgentDispatchTool(o) {
     }
 
     if (toolName === 'ask_user_question') {
+      const previewQs = args.questions
+      if (
+        typeof bareAgentLooksLikePermissionQuestions === 'function' &&
+        bareAgentLooksLikePermissionQuestions(previewQs)
+      ) {
+        return bareAgentJsonResult({
+          ok: false,
+          error: 'execute_instead',
+          hint:
+            'Do not ask the operator whether to create/move/rename/run. Call write_file, move_path, list_directory now. Keep going until the files exist.'
+        })
+      }
       const rows = Array.isArray(args.questions) ? args.questions : []
       if (!rows.length) {
         return bareAgentJsonResult({ ok: false, error: 'questions_required' })
@@ -9282,6 +9372,19 @@ async function bareAgentDispatchTool(o) {
           summary
         })
       }
+      if (
+        typeof bareAgentLooksLikeQuestionDump === 'function' &&
+        bareAgentLooksLikeQuestionDump(summary)
+      ) {
+        appendProgress('task_complete_rejected_questions')
+        return bareAgentJsonResult({
+          ok: false,
+          error: 'task_not_complete',
+          hint:
+            'That is a question for the operator, not a finish. Call write_file / move_path / list_directory and finish the goal. If blocked, return a BLOCKED REPORT (paths, tool errors, what exists, what is missing).',
+          summary
+        })
+      }
       appendProgress('task_complete: ' + summary.slice(0, 200))
       onTaskComplete(summary || '(done)')
       return bareAgentJsonResult({
@@ -9751,16 +9854,44 @@ async function bareAgentDispatchTool(o) {
         typeof bareAgentNormalizeGuestScript === 'function'
           ? bareAgentNormalizeGuestScript(rawCode)
           : rawCode
-      const scriptPath = paths.dir + '/_tmp_agent_run.mjs'
+      const persistPath =
+        (typeof args.path === 'string' && args.path.trim()) ||
+        (typeof args.file_path === 'string' && args.file_path.trim()) ||
+        (typeof args.save_as === 'string' && args.save_as.trim()) ||
+        ''
+      const tempPath = paths.dir + '/_tmp_agent_run.mjs'
+      let scriptPath = tempPath
+      if (
+        persistPath &&
+        persistPath !== tempPath &&
+        persistPath.startsWith('/') &&
+        !persistPath.includes('..')
+      ) {
+        const persist = await bareAgentDispatchTool({
+          ...o,
+          toolName: 'write_file',
+          argsJson: JSON.stringify({ path: persistPath, content: code })
+        })
+        let persistOk = false
+        try {
+          const pj = JSON.parse(persist)
+          persistOk = Boolean(pj && pj.ok)
+        } catch {
+          persistOk = false
+        }
+        if (persistOk) scriptPath = persistPath
+      }
       if (!vfs?.writeFile || !execLine) {
         return bareAgentJsonResult({ ok: false, error: 'vfs or execLine' })
       }
-      appendProgress('run_js_script (' + code.length + ' chars)')
-      const body =
-        typeof ctx.b4a !== 'undefined' && ctx.b4a && typeof ctx.b4a.from === 'function'
-          ? ctx.b4a.from(code)
-          : new TextEncoder().encode(code)
-      await vfs.writeFile(scriptPath, body)
+      appendProgress('run_js_script (' + code.length + ' chars) path=' + scriptPath)
+      if (scriptPath === tempPath) {
+        const body =
+          typeof ctx.b4a !== 'undefined' && ctx.b4a && typeof ctx.b4a.from === 'function'
+            ? ctx.b4a.from(code)
+            : new TextEncoder().encode(code)
+        await vfs.writeFile(scriptPath, body)
+      }
       /** Absolute path → kernel-runner runs .mjs like `./script.mjs` (no host `node` binary). captureExec adds stdout redirect. */
       const cmd = bareAgentShellQuote(scriptPath)
       const r = await captureExec(cmd, 60000)
@@ -9789,8 +9920,29 @@ async function bareAgentDispatchTool(o) {
           stdout_stderr: text
         })
       }
+      const persisted = scriptPath !== tempPath
+      const persistHint = persisted
+        ? 'Script persisted and executed at ' + scriptPath
+        : 'run_js_script only ran a throwaway temp file at ' +
+          tempPath +
+          '. That is NOT a user file. To create a lasting script call write_file({path, content}) then list_directory the destination.'
       return bareAgentJsonResult(
-        r.ok === false ? r : { ok: true, stdout_stderr: r.stdout_stderr }
+        r.ok === false
+          ? {
+              ok: false,
+              error: out.error || 'run_js_script_failed',
+              stdout_stderr: text,
+              ran_from: scriptPath,
+              persisted_to: persisted ? scriptPath : null,
+              hint: persistHint
+            }
+          : {
+              ok: true,
+              stdout_stderr: r.stdout_stderr,
+              ran_from: scriptPath,
+              persisted_to: persisted ? scriptPath : null,
+              hint: persistHint
+            }
       )
     }
 
@@ -10014,26 +10166,118 @@ async function bareAgentDispatchTool(o) {
     }
 
     if (toolName === 'move_path') {
-      const from = typeof args.from_path === 'string' ? args.from_path : ''
-      const to = typeof args.to_path === 'string' ? args.to_path : ''
+      let from =
+        (typeof args.from_path === 'string' && args.from_path.trim()) ||
+        (typeof args.from === 'string' && args.from.trim()) ||
+        (typeof args.src === 'string' && args.src.trim()) ||
+        (typeof args.source === 'string' && args.source.trim()) ||
+        (typeof args.old_path === 'string' && args.old_path.trim()) ||
+        ''
+      let to =
+        (typeof args.to_path === 'string' && args.to_path.trim()) ||
+        (typeof args.to === 'string' && args.to.trim()) ||
+        (typeof args.dest === 'string' && args.dest.trim()) ||
+        (typeof args.destination === 'string' && args.destination.trim()) ||
+        (typeof args.new_path === 'string' && args.new_path.trim()) ||
+        ''
+      if (!from && paths.dir) {
+        const tmp = paths.dir + '/_tmp_agent_run.mjs'
+        try {
+          if (vfs && typeof vfs.readFile === 'function') {
+            const buf = await vfs.readFile(tmp)
+            if (buf) from = tmp
+          }
+        } catch {
+          /* no temp */
+        }
+      }
+      if (!from || !to) {
+        return bareAgentJsonResult({
+          ok: false,
+          error: 'from_and_to_required',
+          hint: 'move_path needs from_path and to_path (absolute or ~/...). To rename the throwaway script, from_path=~/.agent/_tmp_agent_run.mjs to_path=~/test/bareos-cli.js'
+        })
+      }
       if (
         !bareAgentPathAllowedMutate(from, mutateDenyPrefixes) ||
         !bareAgentPathAllowedMutate(to, mutateDenyPrefixes) ||
         from.includes('..') ||
         to.includes('..')
       ) {
-        return bareAgentJsonResult({ ok: false, error: 'path_not_allowed' })
+        return bareAgentJsonResult({ ok: false, error: 'path_not_allowed', from, to })
+      }
+      let finalTo = to
+      try {
+        const st =
+          vfs && typeof vfs.lstat === 'function'
+            ? await vfs.lstat(to)
+            : vfs && typeof vfs.stat === 'function'
+              ? await vfs.stat(to)
+              : null
+        const isDir =
+          st &&
+          ((typeof st.isDirectory === 'function' && st.isDirectory()) ||
+            (st && st.kind === 'dir'))
+        if (isDir) {
+          let base = from.split('/').filter(Boolean).pop() || 'file'
+          if (base === '_tmp_agent_run.mjs') {
+            const named =
+              (typeof args.filename === 'string' && args.filename.trim()) ||
+              (typeof args.name === 'string' && args.name.trim()) ||
+              'cli.js'
+            base = named.replace(/^.*\//, '')
+          }
+          finalTo = to.replace(/\/+$/, '') + '/' + base
+        }
+      } catch {
+        /* dest does not exist yet — treat as file path */
       }
       if (!execLine) {
         return bareAgentJsonResult({ ok: false, error: 'execLine unavailable' })
       }
-      appendProgress('move_path')
+      appendProgress('move_path ' + from + ' -> ' + finalTo)
+      const parent = finalTo.replace(/\/[^/]+$/, '')
+      if (parent && parent !== finalTo && vfs && typeof vfs.mkdir === 'function') {
+        try {
+          await vfs.mkdir(parent, { recursive: true })
+        } catch {
+          /* optional */
+        }
+      }
       const cmd =
-        'mv -- ' + bareAgentShellQuote(from) + ' ' + bareAgentShellQuote(to)
+        'mv -- ' + bareAgentShellQuote(from) + ' ' + bareAgentShellQuote(finalTo)
       const r = await captureExec(cmd, 120000)
-      return bareAgentJsonResult(
-        r.ok === false ? r : { ok: true, stdout_stderr: r.stdout_stderr }
-      )
+      if (r && r.ok === false) {
+        return bareAgentJsonResult({ ...r, from, to: finalTo })
+      }
+      let verified = false
+      try {
+        if (vfs && typeof vfs.lstat === 'function') verified = Boolean(await vfs.lstat(finalTo))
+        else if (vfs && typeof vfs.stat === 'function') verified = Boolean(await vfs.stat(finalTo))
+        else if (vfs && typeof vfs.readFile === 'function') {
+          const got = await vfs.readFile(finalTo)
+          verified = got != null
+        }
+      } catch {
+        verified = false
+      }
+      if (!verified) {
+        return bareAgentJsonResult({
+          ok: false,
+          error: 'move_not_visible',
+          from,
+          to: finalTo,
+          stdout_stderr: r && r.stdout_stderr,
+          hint: 'mv ran but destination is not visible. Retry with absolute paths and list_directory the target folder.'
+        })
+      }
+      return bareAgentJsonResult({
+        ok: true,
+        from,
+        to: finalTo,
+        verified: true,
+        stdout_stderr: r && r.stdout_stderr
+      })
     }
 
     if (toolName === 'delete_path') {
@@ -13777,7 +14021,7 @@ Bare OS by Raven Scott (https://raven-scott.fyi). Repo: https://git.ssh.surf/snx
 WORK POLICY.
 - Keep every explicit requirement in view until it is done, superseded, or blocked. If blocked, say so plainly.
 - Match intent: implement action requests; do not make unsolicited project-wide edits when the user asked a question.
-- For clear, reversible local work, do it now. NEVER ASK for permission.
+- For clear, reversible local work, do it now. NEVER ASK for permission. Never ask "would you like me to". Pick a reasonable name and do it. Keep calling tools until the original request is verified (list_directory / file_stat / stdout). If you cannot finish, call task_complete with a BLOCKED REPORT: goal, every path tried, verbatim tool errors, what exists now, what is missing. Do not stop to ask questions.
 - Claim done, fixed, or tested only when a tool result supports it. Otherwise say what you did not verify.
 - Scope to what was asked. Comments are short and factual. No placeholders. Comments must not substitute for a fix.
 
@@ -13787,13 +14031,13 @@ ACCESS (denylist, not allowlist). You already have full guest admin. You can cre
 - Read any absolute path, including /proc (read_proc_file, runtime_diagnostic_bundle). Use read_file offset/limit for large files.
 - Run every guest command via run_command (command_deny is empty by default). Prefer list_directory, glob_files, and file_stat over ls/find when you only need names. list_bin lists guest /bin utilities (POSIX-in-JS, not GNU). Never dump curl/ip/ls for the user — call run_command and report stdout.
 - Delete and move are enabled. Do not mutate the read-only base system: /bin /etc /boot /lib /usr /share /proc /dev /sys /run.
-- Node is not installed in the guest. Never plan or run node, npm, or npx here. Do not use run_js_script to read files, write files, or run shell — call read_file, write_file, run_command instead. curl/wget/ip/ping are run_command({command:"curl -s https://api.ipify.org"}), never run_js_script. Only use run_js_script for custom guest JS. The kernel already injects ctx and argv: never write const/let/var ctx, never export. Body must be only async function run(ctx, argv) { const vfs = ctx.vfs; ... }. Do not import Node builtins. Never paste a "final correct script" into the user reply — call the tool.
+- Node is not installed in the guest. Never plan or run node, npm, or npx here. Do not use run_js_script to read files, write files, or run shell — call read_file, write_file, run_command instead. curl/wget/ip/ping are run_command({command:"curl -s https://api.ipify.org"}), never run_js_script. run_js_script always writes a throwaway ~/.agent/_tmp_agent_run.mjs — that is NOT a user file. To create a lasting script in ~/test, call write_file({path:"~/test/name.js", content}) (or pass path= to run_js_script to persist). Then list_directory that folder. Only use run_js_script for custom guest JS. The kernel already injects ctx and argv: never write const/let/var ctx, never export. Body must be only async function run(ctx, argv) { const vfs = ctx.vfs; ... }. Do not import Node builtins. Never paste a "final correct script" into the user reply — call the tool.
 - Live kernel: read_proc_file on /proc/bare_os/features (or features.json) and /proc/bare_os/capabilities.json. Man pages and apropos_man are documentation only — never infer what is enabled from them.
 - web_fetch uses the same host allow/deny list as wget/curl.
 - emit_host_notification and request_host_action are enabled by default. emergency_stop_mutations is the kill switch.
 - DISCORD CHANNEL: You have a Direct Message channel with the operator. Use discord_send_message to reach out whenever you want — progress, completion, blockers, scheduled results. Do not wait for them to message first. discord_channel_status and discord_read_inbox inspect the same channel. Only whitelisted users receive DMs.
 - Never print ~/.agent/config.json, API keys, seeds, or vault material.
-- ask_user_question is only for a real product choice the user must make. Never use it (or chat) to request permission.
+- ask_user_question is only for a real product choice the user must make. Never use it (or chat) to request permission. Never offer numbered menus ("Would you like me to: 1. …"). Execute.
 
 CODING LOOP. Multi-step work: todo_write (merge=true). Large unknown surface: enter_plan_mode, write ~/.agent/plan.md, exit_plan_mode, then implement. Plan mode is read-only except that plan file. Do not stop after a plan-only reply — keep calling tools until verified.
 1. Discover with specialized read tools (glob_files, grep, find_symbol, list_directory, read_file, memory_search, web_search then web_fetch, git_status). Call read_skill when a skill matches. Read before you edit. Walk-up AGENTS.md and .grok/skills from cwd are already injected when present.
@@ -14152,6 +14396,18 @@ function bareAgentRefusalNudge() {
   )
 }
 
+/**
+ * Bounce a question-menu / unverified file claim back into the tool loop.
+ */
+function bareAgentGoalDriveNudge(kind) {
+  const k = String(kind || 'continue')
+  return (
+    '[harness] Do not ask the operator questions. Do not stop. kind=' +
+    k +
+    '. run_js_script is throwaway (~/.agent/_tmp_agent_run.mjs) — that is not a user file. Create lasting files with write_file({path, content}). Move/rename with move_path({from_path, to_path}). Then list_directory the destination and only then claim done. If blocked, task_complete with a BLOCKED REPORT: goal, paths tried, verbatim tool errors, what list_directory shows, what is missing. Call the next tool now.'
+  )
+}
+
 /**
  * Recover tool calls the provider streamed as plain text (Hermes / Qwen XML,
  * ```bash fences, Grok/Claude `bash` dumps) instead of `delta_tool_calls`.
@@ -15638,6 +15894,13 @@ async function bareOsRunAgentSession(ctx, argv0, task, runOpts) {
     const maxIterBase = Number(configRef.current.max_iterations) || 64
     let iter = 0
     let instructionDumpNudges = 0
+    let goalDriveNudges = 0
+    let sawFileMutation = false
+    let sawFileVerify = false
+    const driveGoal =
+      typeof bareAgentLooksLikeActionRequest === 'function'
+        ? bareAgentLooksLikeActionRequest(task)
+        : true
 
     for (;;) {
       if (masterAbort.signal.aborted) {
@@ -16422,6 +16685,42 @@ async function bareOsRunAgentSession(ctx, argv0, task, runOpts) {
           await bareAgentSaveHistory(ctx, paths.history, messages)
           continue
         }
+        const questionDump =
+          typeof bareAgentLooksLikeQuestionDump === 'function' &&
+          bareAgentLooksLikeQuestionDump(assistantContent)
+        const unverifiedClaim =
+          typeof bareAgentLooksLikeUnverifiedFileClaim === 'function' &&
+          bareAgentLooksLikeUnverifiedFileClaim(assistantContent)
+        const needsVerify = sawFileMutation && !sawFileVerify
+        const neverStarted = driveGoal && !sawFileMutation && !sawFileVerify
+        const shouldDrive =
+          questionDump ||
+          needsVerify ||
+          (unverifiedClaim && !sawFileVerify) ||
+          neverStarted
+        if (goalDriveNudges < 5 && shouldDrive) {
+          goalDriveNudges += 1
+          const kind =
+            goalDriveNudges >= 5
+              ? 'last_chance_report'
+              : questionDump
+                ? 'questions'
+                : needsVerify
+                  ? 'need_list_directory'
+                  : unverifiedClaim
+                    ? 'unverified_claim'
+                    : 'continue'
+          messages.push({
+            role: 'user',
+            content:
+              typeof bareAgentGoalDriveNudge === 'function'
+                ? bareAgentGoalDriveNudge(kind)
+                : '[harness] Keep going. Call write_file / move_path / list_directory. Do not ask questions.'
+          })
+          appendProgress('goal_drive_nudge n=' + String(goalDriveNudges) + ' kind=' + kind)
+          await bareAgentSaveHistory(ctx, paths.history, messages)
+          continue
+        }
         if (
           bareAgentAutonomousShouldContinue(configRef.current, {
             completed: completed,
@@ -16520,6 +16819,35 @@ async function bareOsRunAgentSession(ctx, argv0, task, runOpts) {
         bareAgentWriteOut(ctx, stdout, spinner + '\r')
         statusLineActive = true
 
+        if (
+          name === 'write_file' ||
+          name === 'create_directory' ||
+          name === 'move_path' ||
+          name === 'copy_path' ||
+          name === 'delete_path' ||
+          name === 'edit_file' ||
+          name === 'search_replace' ||
+          name === 'apply_patch'
+        ) {
+          sawFileMutation = true
+        }
+        if (name === 'run_js_script') {
+          try {
+            const a = JSON.parse(argsStr || '{}')
+            if (a && (a.path || a.file_path || a.save_as)) sawFileMutation = true
+          } catch {
+            /* ignore */
+          }
+        }
+        if (
+          name === 'list_directory' ||
+          name === 'file_stat' ||
+          name === 'read_file' ||
+          name === 'glob_files'
+        ) {
+          sawFileVerify = true
+        }
+
         let resultStr = await bareAgentDispatchTool({
           ctx,
           toolName: name,
@@ -16532,6 +16860,23 @@ async function bareOsRunAgentSession(ctx, argv0, task, runOpts) {
           manCacheRef,
           onTaskComplete
         })
+        if (name === 'task_complete') {
+          let parsedTc = null
+          try {
+            parsedTc = JSON.parse(resultStr)
+          } catch {
+            parsedTc = null
+          }
+          if (parsedTc && parsedTc.ok && sawFileMutation && !sawFileVerify) {
+            completed = false
+            resultStr = JSON.stringify({
+              ok: false,
+              error: 'not_verified',
+              hint:
+                'You created or moved files but never listed them. Call list_directory on the destination, then task_complete with what it shows. If blocked, return a BLOCKED REPORT.'
+            })
+          }
+        }
         if (typeof bareAgentRunHooks === 'function' && paths.hooks) {
           try {
             let parsedArgs = {}