Agent Harness Updates

packages/bare-os-coreutils/test/agent-config-surface.test.mjs

@@ -222,6 +222,8 @@ test('agent-tui embeds operating contract appendix', async (t) => {
   t.ok(TUI.includes('YOU RUN THE TOOLS'))
   t.ok(TUI.includes('YOU CAN EXECUTE'))
   t.ok(TUI.includes('There is no bash, shell, cli, or terminal tool'))
+  t.ok(TUI.includes('never wrap them in run_js_script'))
+  t.ok(TUI.includes('Do not call task_complete after a failed tool'))
   t.ok(TUI.includes('The user never runs your tools'))
   t.ok(TUI.includes('discord_send_message'))
   t.ok(TUI.includes('DISCORD CHANNEL'))

packages/bare-os-coreutils/test/agent-helpers.test.mjs

@@ -29,6 +29,40 @@ test('mutate paths use a denylist (base system read-only, rest writable)', async
   t.absent(s.bareAgentPathAllowedMutate('/'))
 })
 
+test('js script wrapping curl/fetch diverts to a shell command', async (t) => {
+  const s = load()
+  const fromJs = s.bareAgentShellCommandFromJsScript
+  t.is(
+    fromJs("await ctx.execLine('curl http://api.ipify.org')"),
+    'curl http://api.ipify.org'
+  )
+  t.is(
+    fromJs("const r = await fetch('https://api.ipify.org')"),
+    'curl -s https://api.ipify.org'
+  )
+  t.is(fromJs('curl http://api.ipify.org'), 'curl http://api.ipify.org')
+  t.is(fromJs('const vfs = ctx.vfs\nawait vfs.readFile("/tmp/x")'), '')
+})
+
+test('failed task_complete detector matches screenshot surrender', async (t) => {
+  const s = load()
+  const fn = s.bareAgentLooksLikeFailedTaskComplete
+  t.ok(
+    fn(
+      'Done: Attempted to retrieve IP address using JavaScript. Error occurred due to async/await in top-level scope. Alternative method required.'
+    )
+  )
+  t.absent(fn('Fetched public IP via curl; stdout is 1.2.3.4'))
+})
+
+test('guest CLI request detector matches run curl for IP', async (t) => {
+  const s = load()
+  const fn = s.bareAgentLooksLikeGuestCliRequest
+  t.ok(fn('run curl to find our IP Address'))
+  t.ok(fn('get our public ip'))
+  t.absent(fn('refactor the agent prompt'))
+})
+
 test('normalize guest script strips ctx redeclare and export', async (t) => {
   const s = load()
   const n = s.bareAgentNormalizeGuestScript

packages/bare-os-coreutils/test/agent-tool-calls.test.mjs

@@ -177,3 +177,23 @@ test('extract ignores javascript fences that are not shell', () => {
   const out = bareAgentExtractToolCallsFromText(text)
   assert.equal(out.length, 0)
 })
+
+test('extract recovers curl glued to following prose', () => {
+  const { bareAgentExtractToolCallsFromText } = loadToolCallFns()
+  const text =
+    'For example:\ncurl http://api.ipify.orgBut the assistant needs to make sure that the script is correct.'
+  const out = bareAgentExtractToolCallsFromText(text)
+  assert.ok(out.length >= 1)
+  assert.equal(out[0].function.name, 'run_command')
+  assert.equal(JSON.parse(out[0].function.arguments).command, 'curl http://api.ipify.org')
+})
+
+test('third-person planning dump is treated as instruction dump', () => {
+  const { bareAgentLooksLikeInstructionDump } = loadToolCallFns()
+  assert.equal(
+    bareAgentLooksLikeInstructionDump(
+      'But the assistant is supposed to use the run_js_script tool. So the assistant would generate a script that uses curl.'
+    ),
+    true
+  )
+})

packages/bare-os-coreutils/test/agent-tools-run-command.test.mjs

@@ -3,6 +3,8 @@ import { readFileSync } from 'node:fs'
 import vm from 'node:vm'
 
 const CODE =
+  readFileSync(new URL('../lib/agent/agent-helpers.js', import.meta.url), 'utf8') +
+  '\n' +
   readFileSync(new URL('../lib/agent/agent-tools.js', import.meta.url), 'utf8') +
   '\n' +
   readFileSync(
@@ -168,3 +170,41 @@ test('run_command accepts cmd alias field', async (t) => {
   t.ok(seen.length >= 1)
   t.ok(seen[0].startsWith('echo hi > '))
 })
+
+test('run_js_script wrapping curl diverts to run_command', async (t) => {
+  const { ctx, seen } = makeCtx()
+  const out = await callNamed(ctx, 'run_js_script', {
+    code: "await ctx.execLine('curl http://api.ipify.org')"
+  })
+  t.ok(out.ok)
+  t.ok(seen.length >= 1)
+  t.ok(String(seen[0]).includes('curl http://api.ipify.org'))
+})
+
+test('task_complete after a failed attempt is rejected', async (t) => {
+  const { ctx } = makeCtx()
+  let completed = false
+  const s = load()
+  const dispatch = /** @type {(o: object) => Promise<string>} */ (s.bareAgentDispatchTool)
+  const raw = await dispatch({
+    ctx,
+    paths: { cmdOut: '/tmp/agent.out', dir: '/tmp' },
+    toolName: 'task_complete',
+    argsJson: JSON.stringify({
+      summary:
+        'Done: Attempted to retrieve IP address using JavaScript. Error occurred due to async/await in top-level scope. Alternative method required.'
+    }),
+    configRef: { current: {} },
+    appendProgress: () => {},
+    signal: null,
+    onTaskComplete: () => {
+      completed = true
+    },
+    bareWebRunTool: async () => ({ ok: false }),
+    bareWebFmtErr: (e) => String(e)
+  })
+  const out = JSON.parse(raw)
+  t.absent(out.ok)
+  t.is(out.error, 'task_not_complete')
+  t.absent(completed)
+})