From e74664fe9e6fbb4dbfcdfb726e6d51c5b328a9cf Mon Sep 17 00:00:00 2001
From: Joncallim <joncallim@gmail.com>
Date: Thu, 9 Jul 2026 14:50:49 +0000
Subject: [PATCH 1/9] Harden work-package execution repair

---
 web/__tests__/work-package-executor.test.ts | 164 +++++++++++++++++++-
 web/worker/work-package-executor.ts         |  76 +++++++--
 2 files changed, 217 insertions(+), 23 deletions(-)

diff --git a/web/__tests__/work-package-executor.test.ts b/web/__tests__/work-package-executor.test.ts
index e5ab649..5b35963 100644
--- a/web/__tests__/work-package-executor.test.ts
+++ b/web/__tests__/work-package-executor.test.ts
@@ -188,6 +188,34 @@ describe('parseWorkPackageExecutionPlan', () => {
     expect(parsed.summary).toBe('Built tracker')
     expect(parsed.files).toHaveLength(1)
   })
+
+  it('parses a one-line fenced execution JSON block', () => {
+    const payload = JSON.stringify({
+      schemaVersion: 1,
+      summary: 'Built tracker',
+      files: [{ path: 'package.json', content: '{"scripts":{}}' }],
+      commands: [],
+    })
+
+    const parsed = parseWorkPackageExecutionPlan(`\`\`\`work_package_execution_json ${payload}\`\`\``)
+
+    expect(parsed.summary).toBe('Built tracker')
+    expect(parsed.files).toHaveLength(1)
+  })
+
+  it('parses a JSON-encoded execution response string', () => {
+    const payload = JSON.stringify({
+      schemaVersion: 1,
+      summary: 'Built tracker',
+      files: [{ path: 'package.json', content: '{"scripts":{}}' }],
+      commands: [],
+    })
+
+    const parsed = parseWorkPackageExecutionPlan(JSON.stringify(payload))
+
+    expect(parsed.summary).toBe('Built tracker')
+    expect(parsed.files).toHaveLength(1)
+  })
 })
 
 describe('hasLocalConflictCopyPathSegment', () => {
@@ -319,7 +347,7 @@ describe('executeWorkPackage', () => {
           },
           {
             path: 'index.test.js',
-            content: 'import test from "node:test"; import assert from "node:assert/strict"; test("ok", () => assert.equal(1, 1));\n',
+            content: 'const test = require("node:test"); const assert = require("node:assert/strict"); test("ok", () => assert.equal(1, 1));\n',
           },
         ],
         commands: [['npm', 'test'], ['npm', 'run', 'build']],
@@ -1131,7 +1159,7 @@ describe('executeWorkPackage', () => {
           },
           {
             path: 'index.test.js',
-            content: 'import test from "node:test"; import assert from "node:assert/strict"; test("ok", () => assert.equal(1, 1));\n',
+            content: 'const test = require("node:test"); const assert = require("node:assert/strict"); test("ok", () => assert.equal(1, 1));\n',
           },
         ],
         commands: [['npm', 'test']],
@@ -1261,7 +1289,7 @@ describe('executeWorkPackage', () => {
         validationStatus: 'failed',
       })
       await expect(fs.stat(sandbox)).resolves.toMatchObject({})
-      expect(mocks.generateText).toHaveBeenCalledTimes(2)
+      expect(mocks.generateText).toHaveBeenCalledTimes(3)
     }
   })
 
@@ -1523,7 +1551,7 @@ describe('executeWorkPackage', () => {
         prompt: 'Build a tiny task tracker web app. Add focused tests and make sure the app builds.',
       },
     }))).rejects.toThrow(/placeholder tests/i)
-    expect(mocks.generateText).toHaveBeenCalledTimes(2)
+    expect(mocks.generateText).toHaveBeenCalledTimes(3)
   })
 
   it('rejects invalid node:test shell scripts before command execution', async () => {
@@ -1543,7 +1571,7 @@ describe('executeWorkPackage', () => {
           },
           {
             path: 'tracker.test.js',
-            content: 'import test from "node:test"; import assert from "node:assert/strict"; test("adds item", () => assert.equal(1, 1));\n',
+            content: 'const test = require("node:test"); const assert = require("node:assert/strict"); test("adds item", () => assert.equal(1, 1));\n',
           },
           { path: 'build-check.js', content: 'console.log("build ok");\n' },
         ],
@@ -1596,7 +1624,7 @@ describe('executeWorkPackage', () => {
             },
             {
               path: 'tracker.test.js',
-              content: 'import test from "node:test"; import assert from "node:assert/strict"; test("adds item", () => assert.equal(["a"].length, 1));\n',
+              content: 'const test = require("node:test"); const assert = require("node:assert/strict"); test("adds item", () => assert.equal(["a"].length, 1));\n',
             },
             {
               path: 'build-check.js',
@@ -1618,4 +1646,128 @@ describe('executeWorkPackage', () => {
     expect(result.summary).toBe('Built and tested the tracker.')
     expect(result.commandResults.map((item) => item.exitCode)).toEqual([0, 0])
   })
+
+  it('reprompts through invalid JSON and bad generated tests for a tiny task tracker', async () => {
+    mocks.generateText
+      .mockResolvedValueOnce({
+        text: '```work_package_execution_json\n{"schemaVersion":1,"summary":"Cut off"',
+      })
+      .mockResolvedValueOnce({
+        text: JSON.stringify({
+          schemaVersion: 1,
+          summary: 'Generated weak tracker tests.',
+          files: [
+            {
+              path: 'package.json',
+              content: JSON.stringify({
+                scripts: {
+                  build: 'node build-check.js',
+                  test: 'node test.js',
+                },
+              }),
+            },
+            {
+              path: 'tracker.test.js',
+              content: 'const test = require("node:test"); test("adds item", () => console.log("ok"));\n',
+            },
+            {
+              path: 'build-check.js',
+              content: 'console.log("build validated");\n',
+            },
+          ],
+          commands: [['npm', 'test'], ['npm', 'run', 'build']],
+        }),
+      })
+      .mockResolvedValueOnce({
+        text: JSON.stringify({
+          schemaVersion: 1,
+          summary: 'Built a dependency-free tiny task tracker.',
+          files: [
+            {
+              path: 'package.json',
+              content: JSON.stringify({
+                scripts: {
+                  build: 'node build-check.js',
+                  test: 'node --test',
+                },
+              }),
+            },
+            {
+              path: 'taskStore.js',
+              content: [
+                'function createState() { return { tasks: [] }; }',
+                'function addTask(state, title) {',
+                '  const task = { id: String(state.tasks.length + 1), title, completed: false };',
+                '  return { tasks: [...state.tasks, task] };',
+                '}',
+                'function toggleTask(state, id) {',
+                '  return { tasks: state.tasks.map((task) => task.id === id ? { ...task, completed: !task.completed } : task) };',
+                '}',
+                'function deleteTask(state, id) { return { tasks: state.tasks.filter((task) => task.id !== id) }; }',
+                'function filterTasks(state, filter) {',
+                '  if (filter === "active") return state.tasks.filter((task) => !task.completed);',
+                '  if (filter === "completed") return state.tasks.filter((task) => task.completed);',
+                '  return state.tasks;',
+                '}',
+                'function hydrate(raw) {',
+                '  try {',
+                '    const parsed = JSON.parse(raw);',
+                '    return Array.isArray(parsed.tasks) ? parsed : createState();',
+                '  } catch {',
+                '    return createState();',
+                '  }',
+                '}',
+                'module.exports = { createState, addTask, toggleTask, deleteTask, filterTasks, hydrate };',
+                '',
+              ].join('\n'),
+            },
+            {
+              path: 'tracker.test.js',
+              content: [
+                'const test = require("node:test");',
+                'const assert = require("node:assert/strict");',
+                'const { createState, addTask, toggleTask, deleteTask, filterTasks, hydrate } = require("./taskStore.js");',
+                '',
+                'test("add, complete, filter, delete, and malformed storage fallback", () => {',
+                '  let state = createState();',
+                '  state = addTask(state, "Ship the tracker");',
+                '  assert.equal(state.tasks.length, 1);',
+                '  state = toggleTask(state, state.tasks[0].id);',
+                '  assert.deepEqual(filterTasks(state, "completed").map((task) => task.title), ["Ship the tracker"]);',
+                '  assert.equal(filterTasks(state, "active").length, 0);',
+                '  state = deleteTask(state, state.tasks[0].id);',
+                '  assert.equal(filterTasks(state, "all").length, 0);',
+                '  assert.deepEqual(hydrate("{bad json"), createState());',
+                '});',
+                '',
+              ].join('\n'),
+            },
+            {
+              path: 'build-check.js',
+              content: [
+                'const fs = require("node:fs");',
+                'for (const file of ["taskStore.js", "tracker.test.js"]) {',
+                '  if (!fs.existsSync(file)) throw new Error(`${file} is missing`);',
+                '}',
+                '',
+              ].join('\n'),
+            },
+          ],
+          commands: [['npm', 'test'], ['npm', 'run', 'build']],
+        }),
+      })
+
+    const result = await executeWorkPackage(context({
+      task: {
+        ...context().task,
+        prompt: 'Build a tiny task tracker web app. Add focused tests and make sure the app builds.',
+      },
+    }))
+
+    expect(mocks.generateText).toHaveBeenCalledTimes(3)
+    expect(mocks.generateText.mock.calls[1][0].prompt).toContain('Execution response was not valid JSON.')
+    expect(mocks.generateText.mock.calls[2][0].prompt).toContain('Generated test file is not a focused node:test assertion: tracker.test.js')
+    expect(result.summary).toBe('Built a dependency-free tiny task tracker.')
+    expect(result.commandResults.map((item) => item.exitCode)).toEqual([0, 0])
+  })
 })
diff --git a/web/worker/work-package-executor.ts b/web/worker/work-package-executor.ts
index ea5fb09..100087d 100644
--- a/web/worker/work-package-executor.ts
+++ b/web/worker/work-package-executor.ts
@@ -36,7 +36,7 @@ const MAX_FILE_BYTES = 512 * 1024
 const MAX_COMMANDS = 5
 const MAX_COMMAND_OUTPUT_BYTES = 16 * 1024
 const COMMAND_TIMEOUT_MS = 120_000
-const MAX_GENERATION_ATTEMPTS = 2
+const MAX_GENERATION_ATTEMPTS = 3
 const DEFAULT_GENERATION_TIMEOUT_MS = 120_000
 const DEFAULT_GENERATION_MAX_OUTPUT_TOKENS = 8000
 export const MAX_WORK_PACKAGE_EXECUTION_ATTEMPTS = 3
@@ -289,9 +289,26 @@ function assertAllowedCommand(command: string[]): void {
 // small number of candidate start positions is more than enough.
 const MAX_JSON_SCAN_ATTEMPTS = 64
 
-function extractJson(rawText: string): string {
-  const fenced = /```(?:work_package_execution_json|json)?\s*\n([\s\S]*?)\n?```/i.exec(rawText)
-  if (fenced) return fenced[1]
+function uniqueNonEmpty(values: string[]): string[] {
+  const seen = new Set<string>()
+  const result: string[] = []
+  for (const value of values) {
+    const trimmed = value.trim()
+    if (trimmed === '' || seen.has(trimmed)) continue
+    seen.add(trimmed)
+    result.push(trimmed)
+  }
+  return result
+}
+
+function extractJsonCandidates(rawText: string): string[] {
+  const candidates: string[] = []
+  const fencedPattern = /```([a-zA-Z0-9_-]+)?[^\S\r\n]*(?:\r?\n)?([\s\S]*?)```/g
+  for (const match of rawText.matchAll(fencedPattern)) {
+    const language = match[1]?.toLowerCase() ?? ''
+    if (language !== '' && language !== 'json' && language !== 'work_package_execution_json') continue
+    candidates.push(match[2])
+  }
 
   let attempts = 0
   for (let start = rawText.indexOf('{'); start >= 0; start = rawText.indexOf('{', start + 1)) {
@@ -320,13 +337,23 @@ function extractJson(rawText: string): string {
       } else if (char === '}') {
         depth -= 1
         if (depth === 0) {
-          return rawText.slice(start, index + 1)
+          candidates.push(rawText.slice(start, index + 1))
+          break
         }
       }
     }
   }
 
-  return rawText
+  candidates.push(rawText)
+  return uniqueNonEmpty(candidates)
+}
+
+function parseJsonCandidate(candidate: string): unknown {
+  let parsed = JSON.parse(candidate) as unknown
+  if (typeof parsed === 'string' && parsed.trim().startsWith('{')) {
+    parsed = JSON.parse(parsed) as unknown
+  }
+  return parsed
 }
 
 function normalizeExecutionPlan(parsed: unknown): WorkPackageExecutionPlan {
@@ -448,10 +475,8 @@ function validatePlanAgainstPrompt(plan: WorkPackageExecutionPlan, prompt: strin
       throw new Error('The user requested tests, but the execution plan did not run npm test.')
     }
 
-    const hasTestFile = plan.files.some((file) =>
-      /(^|\/)(__tests__\/|.*(\.test|\.spec)\.[cm]?[jt]sx?$|test\.[cm]?js$)/i.test(file.path),
-    )
-    if (!hasTestFile) {
+    const testFiles = plan.files.filter((file) => isJavaScriptFile(file.path) && isTestFile(file.path))
+    if (testFiles.length === 0) {
       throw new Error('The user requested focused tests, but the execution plan did not include a test file.')
     }
 
@@ -465,6 +490,11 @@ function validatePlanAgainstPrompt(plan: WorkPackageExecutionPlan, prompt: strin
     if (isUnsafePackageScript(testScript)) {
       throw new Error('The generated test script includes unsafe shell behavior.')
     }
+    for (const file of testFiles) {
+      if (!/\bnode:test\b/.test(file.content) || !/\bassert\b/.test(file.content) || isPlaceholderContent(file.content)) {
+        throw new Error(`Generated test file is not a focused node:test assertion: ${file.path}`)
+      }
+    }
   }
 
   if (promptRequestsBuild(prompt)) {
@@ -533,6 +563,9 @@ async function generateValidatedExecutionPlan(input: {
         '',
         'Return a corrected full `work_package_execution_json` response.',
         'Do not reuse placeholder tests or echo-only build scripts.',
+        'For dependency-free JavaScript tests, set package.json scripts.test to `node --test`.',
+        'Name test files `*.test.js` and import or require `node:test` plus `node:assert/strict`; assert real requested behavior.',
+        'Do not return `node:test` as a shell command, console-only tests, placeholder tests, or prose outside the JSON fence.',
       ].join('\n')
     }
   }
@@ -541,14 +574,23 @@ async function generateValidatedExecutionPlan(input: {
 }
 
 export function parseWorkPackageExecutionPlan(rawText: string): WorkPackageExecutionPlan {
-  const jsonText = extractJson(rawText)
-  let parsed: unknown
-  try {
-    parsed = JSON.parse(jsonText)
-  } catch {
-    throw new Error('Execution response was not valid JSON.')
+  let parseError: Error | null = null
+  let shapeError: Error | null = null
+  for (const jsonText of extractJsonCandidates(rawText)) {
+    let parsed: unknown
+    try {
+      parsed = parseJsonCandidate(jsonText)
+    } catch {
+      parseError ??= new Error('Execution response was not valid JSON.')
+      continue
+    }
+    try {
+      return normalizeExecutionPlan(parsed)
+    } catch (err) {
+      shapeError ??= err instanceof Error ? err : new Error(String(err))
+    }
   }
-  return normalizeExecutionPlan(parsed)
+  throw shapeError ?? parseError ?? new Error('Execution response was not valid JSON.')
 }
 
 export function hasLocalConflictCopyPathSegment(filePath: string): boolean {

From fc587c88fa1df35e39d2c122e9c64f9ea813d17b Mon Sep 17 00:00:00 2001
From: Joncallim <64296013+Joncallim@users.noreply.github.com>
Date: Fri, 10 Jul 2026 12:19:37 +0800
Subject: [PATCH 2/9] Prioritize execution plan parse feedback

---
 web/__tests__/work-package-executor.test.ts | 10 ++++++++++
 web/worker/work-package-executor.ts         | 14 ++++++++------
 2 files changed, 18 insertions(+), 6 deletions(-)

diff --git a/web/__tests__/work-package-executor.test.ts b/web/__tests__/work-package-executor.test.ts
index 5b35963..e9eed58 100644
--- a/web/__tests__/work-package-executor.test.ts
+++ b/web/__tests__/work-package-executor.test.ts
@@ -189,6 +189,16 @@ describe('parseWorkPackageExecutionPlan', () => {
     expect(parsed.files).toHaveLength(1)
   })
 
+  it('reports malformed execution JSON instead of an incidental object shape error', () => {
+    const raw = [
+      'Diagnostic metadata: {"provider":"local"}',
+      '```work_package_execution_json',
+      '{"schemaVersion":1,"summary":"Cut off"',
+    ].join('\n')
+
+    expect(() => parseWorkPackageExecutionPlan(raw)).toThrow(/not valid JSON/i)
+  })
+
   it('parses a one-line fenced execution JSON block', () => {
     const payload = JSON.stringify({
       schemaVersion: 1,
diff --git a/web/worker/work-package-executor.ts b/web/worker/work-package-executor.ts
index 100087d..7e551a2 100644
--- a/web/worker/work-package-executor.ts
+++ b/web/worker/work-package-executor.ts
@@ -310,6 +310,10 @@ function extractJsonCandidates(rawText: string): string[] {
     candidates.push(match[2])
   }
 
+  // The full response is a stronger error signal than brace-scan fallbacks:
+  // those scans may find unrelated JSON in prose surrounding a malformed plan.
+  candidates.push(rawText)
+
   let attempts = 0
   for (let start = rawText.indexOf('{'); start >= 0; start = rawText.indexOf('{', start + 1)) {
     if (++attempts > MAX_JSON_SCAN_ATTEMPTS) break
@@ -344,7 +348,6 @@ function extractJsonCandidates(rawText: string): string[] {
     }
   }
 
-  candidates.push(rawText)
   return uniqueNonEmpty(candidates)
 }
 
@@ -574,23 +577,22 @@ async function generateValidatedExecutionPlan(input: {
 }
 
 export function parseWorkPackageExecutionPlan(rawText: string): WorkPackageExecutionPlan {
-  let parseError: Error | null = null
-  let shapeError: Error | null = null
+  let firstError: Error | null = null
   for (const jsonText of extractJsonCandidates(rawText)) {
     let parsed: unknown
     try {
       parsed = parseJsonCandidate(jsonText)
     } catch {
-      parseError ??= new Error('Execution response was not valid JSON.')
+      firstError ??= new Error('Execution response was not valid JSON.')
       continue
     }
     try {
       return normalizeExecutionPlan(parsed)
     } catch (err) {
-      shapeError ??= err instanceof Error ? err : new Error(String(err))
+      firstError ??= err instanceof Error ? err : new Error(String(err))
     }
   }
-  throw shapeError ?? parseError ?? new Error('Execution response was not valid JSON.')
+  throw firstError ?? new Error('Execution response was not valid JSON.')
 }
 
 export function hasLocalConflictCopyPathSegment(filePath: string): boolean {

From d24f64aec223db18a511d9694fa17b33e6a676a5 Mon Sep 17 00:00:00 2001
From: Joncallim <64296013+Joncallim@users.noreply.github.com>
Date: Fri, 10 Jul 2026 12:26:16 +0800
Subject: [PATCH 3/9] Close execution repair review gaps

---
 web/__tests__/work-package-executor.test.ts | 96 +++++++++++++++++++++
 web/worker/work-package-executor.ts         | 61 ++++++++++---
 2 files changed, 145 insertions(+), 12 deletions(-)

diff --git a/web/__tests__/work-package-executor.test.ts b/web/__tests__/work-package-executor.test.ts
index e9eed58..0b12023 100644
--- a/web/__tests__/work-package-executor.test.ts
+++ b/web/__tests__/work-package-executor.test.ts
@@ -155,6 +155,25 @@ describe('parseWorkPackageExecutionPlan', () => {
     expect(parsed.commands).toEqual([['npm', 'test']])
   })
 
+  it('normalizes package-script validation command aliases', () => {
+    const parsed = parseWorkPackageExecutionPlan(JSON.stringify({
+      schemaVersion: 1,
+      summary: 'Built tracker',
+      files: [{
+        path: 'package.json',
+        content: JSON.stringify({
+          scripts: {
+            build: 'node build-check.js',
+            test: 'node tracker.test.js',
+          },
+        }),
+      }],
+      commands: [['node', 'tracker.test.js'], ['node', 'build-check.js']],
+    }))
+
+    expect(parsed.commands).toEqual([['npm', 'test'], ['npm', 'run', 'build']])
+  })
+
   it('rejects unsupported commands', () => {
     expect(() => parseWorkPackageExecutionPlan(JSON.stringify({
       schemaVersion: 1,
@@ -1597,6 +1616,83 @@ describe('executeWorkPackage', () => {
     }))).rejects.toThrow(/test script is invalid/i)
   })
 
+  it('accepts focused tests that use a localStorage stub and implementation placeholder text', async () => {
+    mocks.generateText.mockResolvedValue({
+      text: JSON.stringify({
+        schemaVersion: 1,
+        summary: 'Built and tested the tracker.',
+        files: [
+          {
+            path: 'package.json',
+            content: JSON.stringify({
+              scripts: {
+                build: 'node build-check.js',
+                test: 'node --test',
+              },
+            }),
+          },
+          {
+            path: 'tracker.js',
+            content: 'export function configure(input) { input.placeholder = "Add task"; }\n',
+          },
+          {
+            path: 'tracker.test.js',
+            content: [
+              'const test = require("node:test");',
+              'const assert = require("node:assert/strict");',
+              'test("persists with a localStorage stub", () => {',
+              '  const localStorageStub = new Map([["tasks", "[]"]]);',
+              '  assert.equal(localStorageStub.get("tasks"), "[]");',
+              '});',
+              '',
+            ].join('\n'),
+          },
+          {
+            path: 'build-check.js',
+            content: 'console.log("build validated");\n',
+          },
+        ],
+        commands: [['node', '--test'], ['node', 'build-check.js']],
+      }),
+    })
+
+    const result = await executeWorkPackage(context({
+      task: {
+        ...context().task,
+        prompt: 'Build a tiny task tracker web app. Add focused tests and make sure the app builds.',
+      },
+    }))
+
+    expect(result.summary).toBe('Built and tested the tracker.')
+    expect(result.commandResults.map((item) => item.command)).toEqual([
+      ['npm', 'test'],
+      ['npm', 'run', 'build'],
+    ])
+  })
+
+  it('repairs a response truncated at the output limit', async () => {
+    mocks.generateText
+      .mockResolvedValueOnce({
+        finishReason: 'length',
+        text: '{"schemaVersion":1,"summary":"Cut off"',
+      })
+      .mockResolvedValueOnce({
+        text: JSON.stringify({
+          schemaVersion: 1,
+          summary: 'Returned a concise complete plan.',
+          files: [{ path: 'package.json', content: '{}' }],
+          commands: [],
+        }),
+      })
+
+    const result = await executeWorkPackage(context())
+
+    expect(mocks.generateText).toHaveBeenCalledTimes(2)
+    expect(mocks.generateText.mock.calls[1][0].prompt).toContain('configured output limit')
+    expect(mocks.generateText.mock.calls[1][0].prompt).toContain('Keep the response concise')
+    expect(result.summary).toBe('Returned a concise complete plan.')
+  })
+
   it('reprompts once when validation rejects the first generated plan', async () => {
     mocks.generateText
       .mockResolvedValueOnce({
diff --git a/web/worker/work-package-executor.ts b/web/worker/work-package-executor.ts
index 7e551a2..01a7598 100644
--- a/web/worker/work-package-executor.ts
+++ b/web/worker/work-package-executor.ts
@@ -283,6 +283,43 @@ function assertAllowedCommand(command: string[]): void {
   }
 }
 
+function normalizeCommandParts(command: unknown): string[] {
+  if (!Array.isArray(command)) throw new Error('Each command must be a string array.')
+  const normalized = command.map((part) => {
+    if (typeof part !== 'string') throw new Error('Each command part must be a string.')
+    return part.trim()
+  }).filter(Boolean)
+  if (normalized.length === 0) throw new Error('Execution command must be a non-empty string array.')
+  return normalized
+}
+
+function scriptCommandMatches(
+  packageJson: Record<string, unknown> | null,
+  scriptName: string,
+  normalizedCommand: string,
+): boolean {
+  const script = packageScript(packageJson, scriptName)
+  return script !== '' && normalizeCommand(script.split(/\s+/)) === normalizedCommand
+}
+
+function normalizeValidationCommand(command: unknown, packageJson: Record<string, unknown> | null): string[] {
+  const normalized = normalizeCommandParts(command)
+  const normalizedCommand = normalizeCommand(normalized)
+  if (ALLOWED_COMMANDS.has(normalizedCommand)) return normalized
+
+  if (normalizedCommand === 'npm run test') return ['npm', 'test']
+  if (normalizedCommand === 'npm build') return ['npm', 'run', 'build']
+
+  const scriptMatches = [
+    { canonical: ['npm', 'test'], name: 'test' },
+    { canonical: ['npm', 'run', 'build'], name: 'build' },
+    { canonical: ['npm', 'run', 'lint'], name: 'lint' },
+  ].filter(({ name }) => scriptCommandMatches(packageJson, name, normalizedCommand))
+
+  if (scriptMatches.length === 1) return scriptMatches[0].canonical
+  throw new Error(`Command is not allowed: ${normalizedCommand}`)
+}
+
 // Bounds the brace-scan fallback below. Restarting a full inner scan from every
 // `{` is O(n²) on adversarial model output (e.g. thousands of unclosed braces),
 // which can stall the shared worker. Real responses put the object first, so a
@@ -390,16 +427,9 @@ function normalizeExecutionPlan(parsed: unknown): WorkPackageExecutionPlan {
     return { path: filePath, content }
   })
 
+  const packageJson = readGeneratedPackageJson(files)
   const commands = Array.isArray(parsed.commands)
-    ? parsed.commands.map((command) => {
-        if (!Array.isArray(command)) throw new Error('Each command must be a string array.')
-        const normalized = command.map((part) => {
-          if (typeof part !== 'string') throw new Error('Each command part must be a string.')
-          return part.trim()
-        }).filter(Boolean)
-        assertAllowedCommand(normalized)
-        return normalized
-      })
+    ? parsed.commands.map((command) => normalizeValidationCommand(command, packageJson))
     : []
   if (commands.length > MAX_COMMANDS) {
     throw new Error(`Execution response included too many commands; maximum is ${MAX_COMMANDS}.`)
@@ -426,7 +456,7 @@ function hasCommand(commands: string[][], expected: string): boolean {
 }
 
 function isPlaceholderContent(content: string): boolean {
-  return /\b(no tests? needed|not needed for this example|placeholder|todo only|stub)\b/i.test(content)
+  return /\b(no tests? needed|not needed for this example|placeholder|todo only)\b/i.test(content)
 }
 
 function readGeneratedPackageJson(files: WorkPackageExecutionFile[]): Record<string, unknown> | null {
@@ -484,7 +514,11 @@ function validatePlanAgainstPrompt(plan: WorkPackageExecutionPlan, prompt: strin
     }
 
     const testScript = packageScript(packageJson, 'test')
-    if (isNoOpScript(testScript) || plan.files.some((file) => isPlaceholderContent(file.content))) {
+    if (
+      isNoOpScript(testScript) ||
+      isPlaceholderContent(testScript) ||
+      testFiles.some((file) => isPlaceholderContent(file.content))
+    ) {
       throw new Error('The generated test plan appears to contain placeholder tests.')
     }
     if (isInvalidNodeTestScript(testScript)) {
@@ -528,6 +562,7 @@ async function generateValidatedExecutionPlan(input: {
     const controller = new AbortController()
     const timeout = setTimeout(() => controller.abort(), generationTimeoutMs())
     let text: string
+    let responseError: Error | null = null
     try {
       const generated = await generateText({
         abortSignal: controller.signal,
@@ -538,7 +573,7 @@ async function generateValidatedExecutionPlan(input: {
         temperature: 0.1,
       })
       if (generated.finishReason === 'length') {
-        throw new Error(
+        responseError = new Error(
           `Model generation stopped at the configured output limit (${generationMaxOutputTokens()} tokens) before producing a complete execution plan.`,
         )
       }
@@ -553,6 +588,7 @@ async function generateValidatedExecutionPlan(input: {
     }
 
     try {
+      if (responseError) throw responseError
       const plan = parseWorkPackageExecutionPlan(text)
       validatePlanAgainstPrompt(plan, input.taskPrompt)
       return plan
@@ -565,6 +601,7 @@ async function generateValidatedExecutionPlan(input: {
         `Validation error: ${lastError.message}`,
         '',
         'Return a corrected full `work_package_execution_json` response.',
+        'Keep the response concise enough to finish within the configured output limit.',
         'Do not reuse placeholder tests or echo-only build scripts.',
         'For dependency-free JavaScript tests, set package.json scripts.test to `node --test`.',
         'Name test files `*.test.js` and import or require `node:test` plus `node:assert/strict`; assert real requested behavior.',

From 5a3f678c2e308cb4a59ec41343cdfcbd42b7900b Mon Sep 17 00:00:00 2001
From: Joncallim <64296013+Joncallim@users.noreply.github.com>
Date: Fri, 10 Jul 2026 12:36:39 +0800
Subject: [PATCH 4/9] Complete work-package repair hardening

---
 .env.example                                  |   4 +-
 docs/acp-zed-connector.md                     |   9 +-
 ...0006-executable-workforce-beta-boundary.md |   7 +-
 docs/developer-guide.md                       |  10 +-
 docs/operator-guide.md                        |  12 +-
 .../work-package-execution-context.test.ts    |  17 ++-
 web/__tests__/work-package-executor.test.ts   | 116 ++++++++++++++++--
 web/worker/work-package-executor.ts           |  62 ++++++----
 8 files changed, 180 insertions(+), 57 deletions(-)

diff --git a/.env.example b/.env.example
index 1d30853..ff09af1 100644
--- a/.env.example
+++ b/.env.example
@@ -58,8 +58,8 @@ FORGE_WORKER_CLAIM_TIMEOUT_SECONDS=5
 # generated files.
 # FORGE_WORK_PACKAGE_EXECUTION=1
 # FORGE_HOST_REPOSITORY_WRITES=1
-# ACP package execution remains separately disabled by default because ACP
-# adapters are local processes and are not OS-confined by Forge.
+# ACP package execution is enabled after task approval. Set this to 0 when local
+# adapter process access is not acceptable; ACP is not OS-confined by Forge.
 # FORGE_ACP_WORK_PACKAGE_EXECUTION=0
 
 # Workspace root — where Forge stores user-owned operational files: projects,
diff --git a/docs/acp-zed-connector.md b/docs/acp-zed-connector.md
index 5aae6d2..960610e 100644
--- a/docs/acp-zed-connector.md
+++ b/docs/acp-zed-connector.md
@@ -71,10 +71,11 @@ When a task uses an ACP provider, Forge:
 Forge uses one adapter process per call. It does not keep a long-lived pool of
 ACP agents.
 
-Executable Workforce packages do not use ACP providers by default. ACP adapters
-are local processes and Forge does not OS-confine them to the package attempt
-sandbox. Set `FORGE_ACP_WORK_PACKAGE_EXECUTION=1` only after accepting that
-local process risk for the repositories where Forge runs.
+Executable Workforce packages may use a configured ACP provider after task
+approval. ACP adapters are local processes and Forge does not OS-confine them
+to the package attempt sandbox. Set `FORGE_ACP_WORK_PACKAGE_EXECUTION=0` to
+disable ACP package execution on hosts where that local process risk is not
+acceptable.
 
 ## Current Limits
 
diff --git a/docs/adr/0006-executable-workforce-beta-boundary.md b/docs/adr/0006-executable-workforce-beta-boundary.md
index 8aa0f46..1f29f26 100644
--- a/docs/adr/0006-executable-workforce-beta-boundary.md
+++ b/docs/adr/0006-executable-workforce-beta-boundary.md
@@ -28,9 +28,10 @@ edits and a reviewable sandbox copy of every generated file.
   disables model execution and keeps the handoff-artifact-only path.
 - `FORGE_HOST_REPOSITORY_WRITES=0`, `false`, `off`, `no`, or `disabled` lets
   package models run but keeps generated files sandbox-only.
-- ACP-backed work-package execution is separately disabled by default. Set
-  `FORGE_ACP_WORK_PACKAGE_EXECUTION=1` only when the operator accepts that ACP
-  adapters are local processes and are not OS-confined by Forge.
+- ACP-backed work-package execution is enabled after an operator configures an
+  ACP provider and approves the task. ACP adapters are local processes and are
+  not OS-confined by Forge. Set `FORGE_ACP_WORK_PACKAGE_EXECUTION=0`, `false`,
+  `off`, `no`, or `disabled` to prevent ACP package execution.
 - After Architect plan approval, Forge may execute one eligible specialist work
   package at a time. Parallel specialist execution remains out of scope.
 - Forge may collect bounded read-only host-repository context before a package
diff --git a/docs/developer-guide.md b/docs/developer-guide.md
index f997c8d..8e85f86 100644
--- a/docs/developer-guide.md
+++ b/docs/developer-guide.md
@@ -35,10 +35,10 @@ back through the same provider interface used by the worker. The currently wired
 Agent Client Protocol adapters wrap local tools such as Codex CLI and Claude
 Code; the underlying CLI must already be installed, authenticated, and runnable
 on the worker host. Architect ACP calls run in an isolated runtime directory.
-Executable work-package ACP calls are disabled by default because ACP adapters
-are local processes, not OS-confined sandboxes. Operators can opt in with
-`FORGE_ACP_WORK_PACKAGE_EXECUTION=1` after accepting that risk. See [ACP and the
-Zed connector](acp-zed-connector.md).
+Executable work-package ACP calls are enabled after task approval. ACP adapters
+are local processes, not OS-confined sandboxes. Operators can opt out with
+`FORGE_ACP_WORK_PACKAGE_EXECUTION=0` when that risk is not acceptable. See [ACP
+and the Zed connector](acp-zed-connector.md).
 
 ## Local Development
 
@@ -155,7 +155,7 @@ Feature flag defaults:
 | `FORGE_WORK_PACKAGE_HANDOFF` | enabled | Set `0` or `false` to stop package handoff claims. |
 | `FORGE_WORK_PACKAGE_EXECUTION` | enabled | Set `0`, `false`, `off`, `no`, or `disabled` to stop specialist package execution and create handoff artifacts only. |
 | `FORGE_HOST_REPOSITORY_WRITES` | enabled | Set `0`, `false`, `off`, `no`, or `disabled` to keep generated files sandbox-only and skip local project edits. |
-| `FORGE_ACP_WORK_PACKAGE_EXECUTION` | disabled | Set `1`, `true`, `on`, `yes`, or `enabled` only when local ACP package execution is an accepted operator risk. |
+| `FORGE_ACP_WORK_PACKAGE_EXECUTION` | enabled | Set `0`, `false`, `off`, `no`, or `disabled` to block ACP package execution when local adapter process access is not acceptable. |
 | `FORGE_RUNNING_WORK_PACKAGE_STALE_SECONDS` | `900` | Recovery window before a retry marks an interrupted running work package blocked and starts the next eligible attempt. |
 
 ### Executable Workforce Beta
diff --git a/docs/operator-guide.md b/docs/operator-guide.md
index 012e5a5..4f2426b 100644
--- a/docs/operator-guide.md
+++ b/docs/operator-guide.md
@@ -153,11 +153,11 @@ For the currently wired ACP adapters:
 - The adapter wraps the local `codex` or `claude` CLI.
 - The local CLI must already be installed and logged in.
 - The Forge project must have a local folder so Forge can validate and bound
-  repository context. Architect planning uses an isolated runtime directory;
-  executable work-package ACP sessions are disabled by default because local ACP
-  adapters are not OS-confined by Forge. Set
-  `FORGE_ACP_WORK_PACKAGE_EXECUTION=1` only for repositories where that local
-  process access is acceptable.
+  repository context. Architect planning uses an isolated runtime directory.
+  Executable work-package ACP sessions are enabled after task approval, but the
+  local adapter is not OS-confined by Forge. Set
+  `FORGE_ACP_WORK_PACKAGE_EXECUTION=0` where that local process access is not
+  acceptable.
 - Installing the Zed editor is not required; Forge uses Agent Client Protocol
   adapter packages, not the editor itself.
 
@@ -317,7 +317,7 @@ Worker and workspace options:
 | `FORGE_WORK_PACKAGE_HANDOFF` | Set `0` or `false` to disable default work-package handoff claims |
 | `FORGE_WORK_PACKAGE_EXECUTION` | Set `0`, `false`, `off`, `no`, or `disabled` to disable default package execution and create handoff artifacts only |
 | `FORGE_HOST_REPOSITORY_WRITES` | Set `0`, `false`, `off`, `no`, or `disabled` to keep generated files sandbox-only and skip local project edits |
-| `FORGE_ACP_WORK_PACKAGE_EXECUTION` | Set `1`, `true`, `on`, `yes`, or `enabled` only when local ACP package execution is an accepted operator risk |
+| `FORGE_ACP_WORK_PACKAGE_EXECUTION` | Set `0`, `false`, `off`, `no`, or `disabled` to block ACP package execution when local adapter process access is not acceptable |
 | `FORGE_RUNNING_WORK_PACKAGE_STALE_SECONDS` | Defaults to `900`; retry handoff treats older running package rows as interrupted and recovers them before continuing |
 | `FORGE_WORKSPACE_ROOT` | Fixed workspace root override |
 | `FORGE_MCPS_ROOT` | Fixed shared MCP root override |
diff --git a/web/__tests__/work-package-execution-context.test.ts b/web/__tests__/work-package-execution-context.test.ts
index a50fcfe..fdab062 100644
--- a/web/__tests__/work-package-execution-context.test.ts
+++ b/web/__tests__/work-package-execution-context.test.ts
@@ -104,8 +104,10 @@ describe('loadWorkPackageExecutionContext', () => {
     expect(context.validatedProjectRoot).toBe('/workspace/real-project')
   })
 
-  it('blocks ACP-backed executable work packages unless explicitly enabled', async () => {
+  it('blocks ACP-backed executable work packages when explicitly disabled', async () => {
     vi.clearAllMocks()
+    const previous = process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION
+    process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION = '0'
     const project = { id: 'project-1', localPath: '/workspace/project' }
     const task = { id: 'task-1', projectId: 'project-1', pmProviderConfigId: 'provider-task' }
     const workPackage = { id: 'pkg-1', assignedRole: 'backend' }
@@ -116,17 +118,22 @@ describe('loadWorkPackageExecutionContext', () => {
       config: { providerType: 'acp', modelId: 'codex-cli::gpt-5.3-codex-spark' },
     })
 
-    await expect(loadWorkPackageExecutionContext('task-1', 'pkg-1'))
-      .rejects.toThrow(/ACP work-package execution is disabled/i)
+    try {
+      await expect(loadWorkPackageExecutionContext('task-1', 'pkg-1'))
+        .rejects.toThrow(/ACP work-package execution is disabled/i)
+    } finally {
+      if (previous === undefined) delete process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION
+      else process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION = previous
+    }
 
     expect(mocks.assertProjectLocalPathForExecution).not.toHaveBeenCalled()
     expect(mocks.getModel).not.toHaveBeenCalled()
   })
 
-  it('allows ACP-backed executable work packages when ACP execution is explicitly enabled', async () => {
+  it('allows ACP-backed executable work packages by default', async () => {
     vi.clearAllMocks()
     const previous = process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION
-    process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION = '1'
+    delete process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION
     const project = { id: 'project-1', localPath: '/workspace/project' }
     const task = { id: 'task-1', projectId: 'project-1', pmProviderConfigId: 'provider-task' }
     const workPackage = { id: 'pkg-1', assignedRole: 'backend' }
diff --git a/web/__tests__/work-package-executor.test.ts b/web/__tests__/work-package-executor.test.ts
index 0b12023..40e1f9e 100644
--- a/web/__tests__/work-package-executor.test.ts
+++ b/web/__tests__/work-package-executor.test.ts
@@ -245,6 +245,24 @@ describe('parseWorkPackageExecutionPlan', () => {
     expect(parsed.summary).toBe('Built tracker')
     expect(parsed.files).toHaveLength(1)
   })
+
+  it('parses a JSON-encoded fenced execution response string', () => {
+    const payload = [
+      '```work_package_execution_json',
+      JSON.stringify({
+        schemaVersion: 1,
+        summary: 'Built tracker',
+        files: [{ path: 'package.json', content: '{"scripts":{}}' }],
+        commands: [],
+      }),
+      '```',
+    ].join('\n')
+
+    const parsed = parseWorkPackageExecutionPlan(JSON.stringify(payload))
+
+    expect(parsed.summary).toBe('Built tracker')
+    expect(parsed.files).toHaveLength(1)
+  })
 })
 
 describe('hasLocalConflictCopyPathSegment', () => {
@@ -1253,7 +1271,7 @@ describe('executeWorkPackage', () => {
     await expect(executeWorkPackage(context())).rejects.toThrow(/at least one checkable JavaScript source file/i)
   })
 
-  it('throws validation failures with durable sandbox output metadata', async () => {
+  it('wraps command-selected generation validation failures with durable empty sandbox metadata', async () => {
     mocks.generateText.mockResolvedValue({
       text: JSON.stringify({
         schemaVersion: 1,
@@ -1279,16 +1297,15 @@ describe('executeWorkPackage', () => {
       expect(err).toBeInstanceOf(WorkPackageExecutionError)
       const failure = (err as WorkPackageExecutionError).failureDetails
       expect(failure.sandboxPath).toBe(path.join(tempRoot, '.forge', 'task-runs', 'task-1', 'pkg-1', 'attempt-1'))
-      expect(failure.fileCount).toBe(1)
+      expect(failure.fileCount).toBe(0)
       expect(failure.artifactMetadata).toMatchObject({
-        files: ['package.json'],
+        files: [],
         generatedBy: 'work-package-executor',
         repositoryWrites: false,
-        sandboxWrites: true,
+        sandboxWrites: false,
         validationStatus: 'failed',
       })
-      expect(failure.commandResults).toHaveLength(1)
-      expect(failure.commandResults[0].exitCode).not.toBe(0)
+      expect(failure.commandResults).toHaveLength(0)
     }
   })
 
@@ -1642,7 +1659,9 @@ describe('executeWorkPackage', () => {
               'const assert = require("node:assert/strict");',
               'test("persists with a localStorage stub", () => {',
               '  const localStorageStub = new Map([["tasks", "[]"]]);',
+              '  const input = { placeholder: "Add task" };',
               '  assert.equal(localStorageStub.get("tasks"), "[]");',
+              '  assert.equal(input.placeholder, "Add task");',
               '});',
               '',
             ].join('\n'),
@@ -1693,6 +1712,89 @@ describe('executeWorkPackage', () => {
     expect(result.summary).toBe('Returned a concise complete plan.')
   })
 
+  it('repairs selected test validation when the task prompt does not mention tests', async () => {
+    mocks.generateText
+      .mockResolvedValueOnce({
+        text: JSON.stringify({
+          schemaVersion: 1,
+          summary: 'Generated weak tests.',
+          files: [
+            {
+              path: 'package.json',
+              content: JSON.stringify({ scripts: { test: 'node --test' } }),
+            },
+            {
+              path: 'tracker.test.js',
+              content: 'const test = require("node:test"); test("adds", () => console.log("ok"));\n',
+            },
+          ],
+          commands: [['npm', 'test']],
+        }),
+      })
+      .mockResolvedValueOnce({
+        text: JSON.stringify({
+          schemaVersion: 1,
+          summary: 'Generated focused tests.',
+          files: [
+            {
+              path: 'package.json',
+              content: JSON.stringify({ scripts: { test: 'node --test' } }),
+            },
+            {
+              path: 'tracker.test.js',
+              content: 'const test = require("node:test"); const assert = require("node:assert/strict"); test("adds", () => assert.equal(1, 1));\n',
+            },
+          ],
+          commands: [['npm', 'test']],
+        }),
+      })
+
+    const result = await executeWorkPackage(context())
+
+    expect(mocks.generateText).toHaveBeenCalledTimes(2)
+    expect(mocks.generateText.mock.calls[1][0].prompt).toContain('not a focused node:test assertion')
+    expect(result.summary).toBe('Generated focused tests.')
+  })
+
+  it('repairs a selected placeholder build when the task prompt does not mention builds', async () => {
+    mocks.generateText
+      .mockResolvedValueOnce({
+        text: JSON.stringify({
+          schemaVersion: 1,
+          summary: 'Generated a placeholder build.',
+          files: [
+            {
+              path: 'package.json',
+              content: JSON.stringify({ scripts: { build: 'echo "build passed"' } }),
+            },
+            { path: 'app.js', content: 'export const ready = true;\n' },
+          ],
+          commands: [['npm', 'run', 'build']],
+        }),
+      })
+      .mockResolvedValueOnce({
+        text: JSON.stringify({
+          schemaVersion: 1,
+          summary: 'Generated a meaningful build check.',
+          files: [
+            {
+              path: 'package.json',
+              content: JSON.stringify({ scripts: { build: 'node build-check.js --strict true' } }),
+            },
+            { path: 'app.js', content: 'export const ready = true;\n' },
+            { path: 'build-check.js', content: 'console.log("build validated");\n' },
+          ],
+          commands: [['npm', 'run', 'build']],
+        }),
+      })
+
+    const result = await executeWorkPackage(context())
+
+    expect(mocks.generateText).toHaveBeenCalledTimes(2)
+    expect(mocks.generateText.mock.calls[1][0].prompt).toContain('build script appears to be a placeholder')
+    expect(result.summary).toBe('Generated a meaningful build check.')
+  })
+
   it('reprompts once when validation rejects the first generated plan', async () => {
     mocks.generateText
       .mockResolvedValueOnce({
@@ -1774,7 +1876,7 @@ describe('executeWorkPackage', () => {
             },
             {
               path: 'tracker.test.js',
-              content: 'const test = require("node:test"); test("adds item", () => console.log("ok"));\n',
+              content: 'const test = require("node:test"); const assert = require("node:assert/strict"); test("adds item", () => console.log("ok"));\n',
             },
             {
               path: 'build-check.js',
diff --git a/web/worker/work-package-executor.ts b/web/worker/work-package-executor.ts
index 01a7598..e912c8d 100644
--- a/web/worker/work-package-executor.ts
+++ b/web/worker/work-package-executor.ts
@@ -27,6 +27,7 @@ import { sanitizeWorkerMessage } from './redaction'
 import { publishTaskEvent } from './events'
 import { recordTaskLogBestEffort } from './task-logs'
 import { shouldApplyHostRepositoryWrites } from './repository-edit-policy'
+import { defaultOnFeatureFlagEnabled } from './feature-flags'
 
 const execFile = promisify(execFileCallback)
 
@@ -47,8 +48,6 @@ const ALLOWED_COMMANDS = new Set([
   'npm run lint',
 ])
 
-const ACP_EXECUTION_ENABLED_VALUES = new Set(['1', 'true', 'on', 'yes', 'enabled'])
-
 type TaskRow = typeof tasks.$inferSelect
 type ProjectRow = typeof projects.$inferSelect
 type WorkPackageRow = typeof workPackages.$inferSelect
@@ -256,8 +255,7 @@ function isAcpModel(model: LanguageModel): boolean {
 }
 
 function isAcpWorkPackageExecutionEnabled(env: Record<string, string | undefined> = process.env): boolean {
-  const raw = env.FORGE_ACP_WORK_PACKAGE_EXECUTION?.trim().toLowerCase()
-  return raw !== undefined && ACP_EXECUTION_ENABLED_VALUES.has(raw)
+  return defaultOnFeatureFlagEnabled(env.FORGE_ACP_WORK_PACKAGE_EXECUTION)
 }
 
 function generationTimeoutMs(): number {
@@ -325,6 +323,7 @@ function normalizeValidationCommand(command: unknown, packageJson: Record<string
 // which can stall the shared worker. Real responses put the object first, so a
 // small number of candidate start positions is more than enough.
 const MAX_JSON_SCAN_ATTEMPTS = 64
+const MAX_JSON_STRING_DECODE_DEPTH = 2
 
 function uniqueNonEmpty(values: string[]): string[] {
   const seen = new Set<string>()
@@ -389,11 +388,7 @@ function extractJsonCandidates(rawText: string): string[] {
 }
 
 function parseJsonCandidate(candidate: string): unknown {
-  let parsed = JSON.parse(candidate) as unknown
-  if (typeof parsed === 'string' && parsed.trim().startsWith('{')) {
-    parsed = JSON.parse(parsed) as unknown
-  }
-  return parsed
+  return JSON.parse(candidate) as unknown
 }
 
 function normalizeExecutionPlan(parsed: unknown): WorkPackageExecutionPlan {
@@ -456,7 +451,7 @@ function hasCommand(commands: string[][], expected: string): boolean {
 }
 
 function isPlaceholderContent(content: string): boolean {
-  return /\b(no tests? needed|not needed for this example|placeholder|todo only)\b/i.test(content)
+  return /\b(no tests? needed|not needed for this example|todo only)\b/i.test(content)
 }
 
 function readGeneratedPackageJson(files: WorkPackageExecutionFile[]): Record<string, unknown> | null {
@@ -481,11 +476,17 @@ function packageScript(packageJson: Record<string, unknown> | null, name: string
 function isNoOpScript(script: string): boolean {
   return (
     script === '' ||
-    /(^|\s)(true|exit 0)(\s|$)/i.test(script) ||
-    /^\s*echo\s+['"]?(build script not needed|no tests? needed|not needed for this example|ok|done)/i.test(script)
+    /^\s*(?:true|exit\s+0)\s*$/i.test(script) ||
+    /^\s*echo(?:\s|$)/i.test(script)
   )
 }
 
+function isFocusedNodeTestAssertion(content: string): boolean {
+  const hasTestCall = /\b(?:test|it)(?:\.(?:only|skip|todo))?\s*\(/.test(content)
+  const hasAssertionCall = /\b(?:assert(?:\.[A-Za-z_$][\w$]*)?|ok|equal|strictEqual|deepEqual|deepStrictEqual|throws|rejects)\s*\(/.test(content)
+  return /\bnode:test\b/.test(content) && /\bnode:assert\/strict\b/.test(content) && hasTestCall && hasAssertionCall
+}
+
 function isInvalidNodeTestScript(script: string): boolean {
   return /^\s*node:test(\s|$)/i.test(script)
 }
@@ -502,15 +503,17 @@ function isUnsafePackageScript(script: string): boolean {
 
 function validatePlanAgainstPrompt(plan: WorkPackageExecutionPlan, prompt: string): void {
   const packageJson = readGeneratedPackageJson(plan.files)
+  const testsRequested = promptRequestsTests(prompt)
+  const testCommandSelected = hasCommand(plan.commands, 'npm test')
 
-  if (promptRequestsTests(prompt)) {
-    if (!hasCommand(plan.commands, 'npm test')) {
-      throw new Error('The user requested tests, but the execution plan did not run npm test.')
-    }
+  if (testsRequested && !testCommandSelected) {
+    throw new Error('The user requested tests, but the execution plan did not run npm test.')
+  }
 
+  if (testsRequested || testCommandSelected) {
     const testFiles = plan.files.filter((file) => isJavaScriptFile(file.path) && isTestFile(file.path))
     if (testFiles.length === 0) {
-      throw new Error('The user requested focused tests, but the execution plan did not include a test file.')
+      throw new Error('The execution plan selected focused tests but did not include a test file.')
     }
 
     const testScript = packageScript(packageJson, 'test')
@@ -528,17 +531,19 @@ function validatePlanAgainstPrompt(plan: WorkPackageExecutionPlan, prompt: strin
       throw new Error('The generated test script includes unsafe shell behavior.')
     }
     for (const file of testFiles) {
-      if (!/\bnode:test\b/.test(file.content) || !/\bassert\b/.test(file.content) || isPlaceholderContent(file.content)) {
+      if (!isFocusedNodeTestAssertion(file.content) || isPlaceholderContent(file.content)) {
         throw new Error(`Generated test file is not a focused node:test assertion: ${file.path}`)
       }
     }
   }
 
-  if (promptRequestsBuild(prompt)) {
-    if (!hasCommand(plan.commands, 'npm run build')) {
-      throw new Error('The user requested a build check, but the execution plan did not run npm run build.')
-    }
+  const buildRequested = promptRequestsBuild(prompt)
+  const buildCommandSelected = hasCommand(plan.commands, 'npm run build')
+  if (buildRequested && !buildCommandSelected) {
+    throw new Error('The user requested a build check, but the execution plan did not run npm run build.')
+  }
 
+  if (buildRequested || buildCommandSelected) {
     const buildScript = packageScript(packageJson, 'build')
     if (isNoOpScript(buildScript)) {
       throw new Error('The generated build script appears to be a placeholder.')
@@ -615,7 +620,9 @@ async function generateValidatedExecutionPlan(input: {
 
 export function parseWorkPackageExecutionPlan(rawText: string): WorkPackageExecutionPlan {
   let firstError: Error | null = null
-  for (const jsonText of extractJsonCandidates(rawText)) {
+  const candidates = extractJsonCandidates(rawText).map((text) => ({ depth: 0, text }))
+  for (let index = 0; index < candidates.length; index += 1) {
+    const { depth, text: jsonText } = candidates[index]
     let parsed: unknown
     try {
       parsed = parseJsonCandidate(jsonText)
@@ -623,6 +630,11 @@ export function parseWorkPackageExecutionPlan(rawText: string): WorkPackageExecu
       firstError ??= new Error('Execution response was not valid JSON.')
       continue
     }
+    if (typeof parsed === 'string' && depth < MAX_JSON_STRING_DECODE_DEPTH) {
+      const decoded = extractJsonCandidates(parsed).map((text) => ({ depth: depth + 1, text }))
+      candidates.splice(index + 1, 0, ...decoded)
+      continue
+    }
     try {
       return normalizeExecutionPlan(parsed)
     } catch (err) {
@@ -801,7 +813,7 @@ async function validateGeneratedCommand(projectRoot: string, command: string[]):
     for (const file of testFiles) {
       const absolute = path.join(projectRoot, file)
       const content = await fs.readFile(absolute, 'utf8')
-      if (!/\bnode:test\b/.test(content) || !/\bassert\b/.test(content) || isPlaceholderContent(content)) {
+      if (!isFocusedNodeTestAssertion(content) || isPlaceholderContent(content)) {
         throw new Error(`Generated test file is not a focused node:test assertion: ${file}`)
       }
       await safeSyntaxCheck(absolute)
@@ -1511,7 +1523,7 @@ export async function loadWorkPackageExecutionContext(
   if (!provider) throw new Error(`Provider config ${providerConfigId} is missing or inactive.`)
   if (provider.config.providerType === 'acp' && !isAcpWorkPackageExecutionEnabled()) {
     throw new Error(
-      'ACP work-package execution is disabled. Set FORGE_ACP_WORK_PACKAGE_EXECUTION=1 only after accepting that ACP adapters are local processes and are not OS-confined by Forge.',
+      'ACP work-package execution is disabled by FORGE_ACP_WORK_PACKAGE_EXECUTION. Remove the setting or set it to 1 after accepting that ACP adapters are local processes and are not OS-confined by Forge.',
     )
   }
 

From 3818b08ec070c541b7dfb6b013e532c00fe8a100 Mon Sep 17 00:00:00 2001
From: Joncallim <64296013+Joncallim@users.noreply.github.com>
Date: Fri, 10 Jul 2026 12:40:36 +0800
Subject: [PATCH 5/9] Repair selected lint validation

---
 web/__tests__/work-package-executor.test.ts | 59 ++++++++++++++++++++-
 web/worker/work-package-executor.ts         | 13 +++++
 2 files changed, 70 insertions(+), 2 deletions(-)

diff --git a/web/__tests__/work-package-executor.test.ts b/web/__tests__/work-package-executor.test.ts
index 40e1f9e..f6c9c06 100644
--- a/web/__tests__/work-package-executor.test.ts
+++ b/web/__tests__/work-package-executor.test.ts
@@ -164,14 +164,15 @@ describe('parseWorkPackageExecutionPlan', () => {
         content: JSON.stringify({
           scripts: {
             build: 'node build-check.js',
+            lint: 'node lint-check.js',
             test: 'node tracker.test.js',
           },
         }),
       }],
-      commands: [['node', 'tracker.test.js'], ['node', 'build-check.js']],
+      commands: [['node', 'tracker.test.js'], ['node', 'build-check.js'], ['node', 'lint-check.js']],
     }))
 
-    expect(parsed.commands).toEqual([['npm', 'test'], ['npm', 'run', 'build']])
+    expect(parsed.commands).toEqual([['npm', 'test'], ['npm', 'run', 'build'], ['npm', 'run', 'lint']])
   })
 
   it('rejects unsupported commands', () => {
@@ -1795,6 +1796,60 @@ describe('executeWorkPackage', () => {
     expect(result.summary).toBe('Generated a meaningful build check.')
   })
 
+  it('repairs selected lint validation before writing sandbox files', async () => {
+    mocks.generateText
+      .mockResolvedValueOnce({
+        text: JSON.stringify({
+          schemaVersion: 1,
+          summary: 'Generated a placeholder lint command.',
+          files: [
+            {
+              path: 'package.json',
+              content: JSON.stringify({ scripts: { lint: 'echo "lint passed"' } }),
+            },
+            { path: 'app.js', content: 'module.exports = { ready: true };\n' },
+          ],
+          commands: [['npm', 'run', 'lint']],
+        }),
+      })
+      .mockResolvedValueOnce({
+        text: JSON.stringify({
+          schemaVersion: 1,
+          summary: 'Generated unchecked TypeScript lint input.',
+          files: [
+            {
+              path: 'package.json',
+              content: JSON.stringify({ scripts: { lint: 'node lint-check.js' } }),
+            },
+            { path: 'app.tsx', content: 'export const App = () => <div />;\n' },
+          ],
+          commands: [['npm', 'run', 'lint']],
+        }),
+      })
+      .mockResolvedValueOnce({
+        text: JSON.stringify({
+          schemaVersion: 1,
+          summary: 'Generated checkable lint input.',
+          files: [
+            {
+              path: 'package.json',
+              content: JSON.stringify({ scripts: { lint: 'node lint-check.js' } }),
+            },
+            { path: 'app.js', content: 'module.exports = { ready: true };\n' },
+            { path: 'lint-check.js', content: 'console.log("lint validated");\n' },
+          ],
+          commands: [['npm', 'run', 'lint']],
+        }),
+      })
+
+    const result = await executeWorkPackage(context())
+
+    expect(mocks.generateText).toHaveBeenCalledTimes(3)
+    expect(mocks.generateText.mock.calls[1][0].prompt).toContain('lint script appears to be a placeholder')
+    expect(mocks.generateText.mock.calls[2][0].prompt).toContain('at least one checkable JavaScript source file')
+    expect(result.summary).toBe('Generated checkable lint input.')
+  })
+
   it('reprompts once when validation rejects the first generated plan', async () => {
     mocks.generateText
       .mockResolvedValueOnce({
diff --git a/web/worker/work-package-executor.ts b/web/worker/work-package-executor.ts
index e912c8d..6f42f57 100644
--- a/web/worker/work-package-executor.ts
+++ b/web/worker/work-package-executor.ts
@@ -552,6 +552,19 @@ function validatePlanAgainstPrompt(plan: WorkPackageExecutionPlan, prompt: strin
       throw new Error('The generated build script includes unsafe shell behavior.')
     }
   }
+
+  if (hasCommand(plan.commands, 'npm run lint')) {
+    const lintScript = packageScript(packageJson, 'lint')
+    if (isNoOpScript(lintScript)) {
+      throw new Error('The generated lint script appears to be a placeholder.')
+    }
+    if (isUnsafePackageScript(lintScript)) {
+      throw new Error('The generated lint script includes unsafe shell behavior.')
+    }
+    if (!plan.files.some((file) => isJavaScriptFile(file.path))) {
+      throw new Error('Static lint validation requires at least one checkable JavaScript source file.')
+    }
+  }
 }
 
 async function generateValidatedExecutionPlan(input: {

From 504ee9dc496693b99a86e29c1801cbbfbc11e0fd Mon Sep 17 00:00:00 2001
From: Joncallim <64296013+Joncallim@users.noreply.github.com>
Date: Fri, 10 Jul 2026 15:47:46 +0800
Subject: [PATCH 6/9] Close final execution repair review gaps

---
 .../work-package-execution-context.test.ts    |  51 +++++----
 web/__tests__/work-package-executor.test.ts   | 106 ++++++++++++++++--
 web/worker/work-package-executor.ts           | 106 +++++++++++++++++-
 3 files changed, 224 insertions(+), 39 deletions(-)

diff --git a/web/__tests__/work-package-execution-context.test.ts b/web/__tests__/work-package-execution-context.test.ts
index fdab062..19fae5e 100644
--- a/web/__tests__/work-package-execution-context.test.ts
+++ b/web/__tests__/work-package-execution-context.test.ts
@@ -104,31 +104,34 @@ describe('loadWorkPackageExecutionContext', () => {
     expect(context.validatedProjectRoot).toBe('/workspace/real-project')
   })
 
-  it('blocks ACP-backed executable work packages when explicitly disabled', async () => {
-    vi.clearAllMocks()
-    const previous = process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION
-    process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION = '0'
-    const project = { id: 'project-1', localPath: '/workspace/project' }
-    const task = { id: 'task-1', projectId: 'project-1', pmProviderConfigId: 'provider-task' }
-    const workPackage = { id: 'pkg-1', assignedRole: 'backend' }
-    mocks.dbSelect
-      .mockReturnValueOnce(chain([{ task, project, workPackage }]))
-      .mockReturnValueOnce(chain([{ id: 'agent-backend', providerConfigId: null }]))
-    mocks.getProvider.mockResolvedValue({
-      config: { providerType: 'acp', modelId: 'codex-cli::gpt-5.3-codex-spark' },
-    })
-
-    try {
-      await expect(loadWorkPackageExecutionContext('task-1', 'pkg-1'))
-        .rejects.toThrow(/ACP work-package execution is disabled/i)
-    } finally {
-      if (previous === undefined) delete process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION
-      else process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION = previous
-    }
+  it.each(['0', 'flase'])(
+    'blocks ACP-backed executable work packages when the setting is %s',
+    async (setting) => {
+      vi.clearAllMocks()
+      const previous = process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION
+      process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION = setting
+      const project = { id: 'project-1', localPath: '/workspace/project' }
+      const task = { id: 'task-1', projectId: 'project-1', pmProviderConfigId: 'provider-task' }
+      const workPackage = { id: 'pkg-1', assignedRole: 'backend' }
+      mocks.dbSelect
+        .mockReturnValueOnce(chain([{ task, project, workPackage }]))
+        .mockReturnValueOnce(chain([{ id: 'agent-backend', providerConfigId: null }]))
+      mocks.getProvider.mockResolvedValue({
+        config: { providerType: 'acp', modelId: 'codex-cli::gpt-5.3-codex-spark' },
+      })
+
+      try {
+        await expect(loadWorkPackageExecutionContext('task-1', 'pkg-1'))
+          .rejects.toThrow(/ACP work-package execution is disabled/i)
+      } finally {
+        if (previous === undefined) delete process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION
+        else process.env.FORGE_ACP_WORK_PACKAGE_EXECUTION = previous
+      }
 
-    expect(mocks.assertProjectLocalPathForExecution).not.toHaveBeenCalled()
-    expect(mocks.getModel).not.toHaveBeenCalled()
-  })
+      expect(mocks.assertProjectLocalPathForExecution).not.toHaveBeenCalled()
+      expect(mocks.getModel).not.toHaveBeenCalled()
+    },
+  )
 
   it('allows ACP-backed executable work packages by default', async () => {
     vi.clearAllMocks()
diff --git a/web/__tests__/work-package-executor.test.ts b/web/__tests__/work-package-executor.test.ts
index f6c9c06..429b8e1 100644
--- a/web/__tests__/work-package-executor.test.ts
+++ b/web/__tests__/work-package-executor.test.ts
@@ -1223,31 +1223,117 @@ describe('executeWorkPackage', () => {
     await expect(fs.stat(outsideFile)).rejects.toMatchObject({ code: 'ENOENT' })
   })
 
-  it('fails build validation when no JavaScript source files can be checked', async () => {
+  it('repairs build validation when no JavaScript source files can be checked', async () => {
+    mocks.generateText
+      .mockResolvedValueOnce({
+        text: JSON.stringify({
+          schemaVersion: 1,
+          summary: 'Generated unchecked TypeScript.',
+          files: [
+            {
+              path: 'package.json',
+              content: JSON.stringify({ scripts: { build: 'tsc --noEmit' } }),
+            },
+            {
+              path: 'src/app.tsx',
+              content: 'export const App = () => <div />\n',
+            },
+          ],
+          commands: [['npm', 'run', 'build']],
+        }),
+      })
+      .mockResolvedValueOnce({
+        text: JSON.stringify({
+          schemaVersion: 1,
+          summary: 'Generated checkable JavaScript.',
+          files: [
+            {
+              path: 'package.json',
+              content: JSON.stringify({ scripts: { build: 'node build-check.js' } }),
+            },
+            { path: 'app.js', content: 'module.exports = { ready: true };\n' },
+            { path: 'build-check.js', content: 'console.log("build validated");\n' },
+          ],
+          commands: [['npm', 'run', 'build']],
+        }),
+      })
+
+    const result = await executeWorkPackage(context({
+      task: {
+        ...context().task,
+        prompt: 'Build a tiny task tracker web app. Make sure it builds.',
+      },
+    }))
+
+    expect(mocks.generateText).toHaveBeenCalledTimes(2)
+    expect(mocks.generateText.mock.calls[1][0].prompt).toContain('at least one checkable JavaScript source file')
+    expect(result.summary).toBe('Generated checkable JavaScript.')
+  })
+
+  it.each([
+    [
+      'comment-only calls',
+      [
+        'const test = require("node:test");',
+        'const assert = require("node:assert/strict");',
+        '// test("fake", () => assert.equal(1, 1));',
+        '',
+      ].join('\n'),
+    ],
+    [
+      'skipped and TODO tests',
+      [
+        'const test = require("node:test");',
+        'const assert = require("node:assert/strict");',
+        'test.skip("skipped", () => assert.equal(1, 1));',
+        'test.todo("todo");',
+        '',
+      ].join('\n'),
+    ],
+    [
+      'string-only calls',
+      [
+        'const test = require("node:test");',
+        'const assert = require("node:assert/strict");',
+        'const example = "test(\\"fake\\", () => assert.equal(1, 1))";',
+        'void example;',
+        '',
+      ].join('\n'),
+    ],
+    [
+      'string-only module references',
+      [
+        'const modules = `require("node:test") require("node:assert/strict")`;',
+        'const test = (_name, callback) => callback();',
+        'const assert = { equal: () => undefined };',
+        'test("fake", () => assert.equal(1, 1));',
+        'void modules;',
+        '',
+      ].join('\n'),
+    ],
+  ])('rejects %s as focused test coverage', async (_label, content) => {
     mocks.generateText.mockResolvedValue({
       text: JSON.stringify({
         schemaVersion: 1,
-        summary: 'Generated unchecked TypeScript.',
+        summary: 'Generated non-running tests.',
         files: [
           {
             path: 'package.json',
-            content: JSON.stringify({ scripts: { build: 'tsc --noEmit' } }),
-          },
-          {
-            path: 'src/app.tsx',
-            content: 'export const App = () => <div />\n',
+            content: JSON.stringify({ scripts: { test: 'node --test' } }),
           },
+          { path: 'tracker.test.js', content },
         ],
-        commands: [['npm', 'run', 'build']],
+        commands: [['npm', 'test']],
       }),
     })
 
     await expect(executeWorkPackage(context({
       task: {
         ...context().task,
-        prompt: 'Build a tiny task tracker web app. Make sure it builds.',
+        prompt: 'Build a tiny task tracker web app with focused tests.',
       },
-    }))).rejects.toThrow(/at least one checkable JavaScript source file/i)
+    }))).rejects.toThrow(/not a focused node:test assertion/i)
+    expect(mocks.generateText).toHaveBeenCalledTimes(3)
   })
 
   it('fails lint validation when no JavaScript source files can be checked', async () => {
diff --git a/web/worker/work-package-executor.ts b/web/worker/work-package-executor.ts
index 6f42f57..a75735a 100644
--- a/web/worker/work-package-executor.ts
+++ b/web/worker/work-package-executor.ts
@@ -27,7 +27,7 @@ import { sanitizeWorkerMessage } from './redaction'
 import { publishTaskEvent } from './events'
 import { recordTaskLogBestEffort } from './task-logs'
 import { shouldApplyHostRepositoryWrites } from './repository-edit-policy'
-import { defaultOnFeatureFlagEnabled } from './feature-flags'
+import { defaultOnFeatureFlagState } from './feature-flags'
 
 const execFile = promisify(execFileCallback)
 
@@ -255,7 +255,8 @@ function isAcpModel(model: LanguageModel): boolean {
 }
 
 function isAcpWorkPackageExecutionEnabled(env: Record<string, string | undefined> = process.env): boolean {
-  return defaultOnFeatureFlagEnabled(env.FORGE_ACP_WORK_PACKAGE_EXECUTION)
+  const state = defaultOnFeatureFlagState(env.FORGE_ACP_WORK_PACKAGE_EXECUTION)
+  return state.recognized && state.enabled
 }
 
 function generationTimeoutMs(): number {
@@ -481,10 +482,102 @@ function isNoOpScript(script: string): boolean {
   )
 }
 
+function scannedJavaScriptText(content: string, stripStrings: boolean): string {
+  const output = content.split('')
+  let state: 'code' | 'single' | 'double' | 'template' | 'line-comment' | 'block-comment' = 'code'
+  let escaped = false
+
+  const blank = (index: number) => {
+    if (output[index] !== '\n' && output[index] !== '\r') output[index] = ' '
+  }
+
+  for (let index = 0; index < content.length; index += 1) {
+    const char = content[index]
+    const next = content[index + 1]
+
+    if (state === 'code') {
+      if (char === '/' && next === '/') {
+        blank(index)
+        blank(index + 1)
+        index += 1
+        state = 'line-comment'
+      } else if (char === '/' && next === '*') {
+        blank(index)
+        blank(index + 1)
+        index += 1
+        state = 'block-comment'
+      } else if (char === "'") {
+        if (stripStrings) blank(index)
+        state = 'single'
+      } else if (char === '"') {
+        if (stripStrings) blank(index)
+        state = 'double'
+      } else if (char === '`') {
+        if (stripStrings) blank(index)
+        state = 'template'
+      }
+      continue
+    }
+
+    if (state === 'line-comment') {
+      if (char === '\n' || char === '\r') state = 'code'
+      else blank(index)
+      continue
+    }
+
+    if (state === 'block-comment') {
+      if (char === '*' && next === '/') {
+        blank(index)
+        blank(index + 1)
+        index += 1
+        state = 'code'
+      } else {
+        blank(index)
+      }
+      continue
+    }
+
+    if (stripStrings) blank(index)
+    if (escaped) {
+      escaped = false
+    } else if (char === '\\') {
+      escaped = true
+    } else if (
+      (state === 'single' && char === "'") ||
+      (state === 'double' && char === '"') ||
+      (state === 'template' && char === '`')
+    ) {
+      state = 'code'
+    }
+  }
+
+  return output.join('')
+}
+
+function hasExecutableModuleReference(commentFree: string, executable: string, pattern: RegExp): boolean {
+  return [...commentFree.matchAll(pattern)].some((match) => {
+    const start = match.index ?? 0
+    const executableMatch = executable.slice(start, start + match[0].length)
+    return /\b(?:require|from|import)\b/.test(executableMatch)
+  })
+}
+
 function isFocusedNodeTestAssertion(content: string): boolean {
-  const hasTestCall = /\b(?:test|it)(?:\.(?:only|skip|todo))?\s*\(/.test(content)
-  const hasAssertionCall = /\b(?:assert(?:\.[A-Za-z_$][\w$]*)?|ok|equal|strictEqual|deepEqual|deepStrictEqual|throws|rejects)\s*\(/.test(content)
-  return /\bnode:test\b/.test(content) && /\bnode:assert\/strict\b/.test(content) && hasTestCall && hasAssertionCall
+  const commentFree = scannedJavaScriptText(content, false)
+  const executable = scannedJavaScriptText(content, true)
+  const hasNodeTestImport = hasExecutableModuleReference(
+    commentFree,
+    executable,
+    /\brequire\s*\(\s*['"]node:test['"]\s*\)|\bfrom\s*['"]node:test['"]|\bimport\s*['"]node:test['"]/g,
+  )
+  const hasNodeAssertImport = hasExecutableModuleReference(
+    commentFree,
+    executable,
+    /\brequire\s*\(\s*['"]node:assert\/strict['"]\s*\)|\bfrom\s*['"]node:assert\/strict['"]|\bimport\s*['"]node:assert\/strict['"]/g,
+  )
+  const hasTestCall = /\b(?:test|it)(?:\.only)?\s*\(/.test(executable)
+  const hasAssertionCall = /\b(?:assert(?:\.[A-Za-z_$][\w$]*)?|ok|equal|strictEqual|deepEqual|deepStrictEqual|throws|rejects)\s*\(/.test(executable)
+  return hasNodeTestImport && hasNodeAssertImport && hasTestCall && hasAssertionCall
 }
 
 function isInvalidNodeTestScript(script: string): boolean {
@@ -551,6 +644,9 @@ function validatePlanAgainstPrompt(plan: WorkPackageExecutionPlan, prompt: strin
     if (isUnsafePackageScript(buildScript)) {
       throw new Error('The generated build script includes unsafe shell behavior.')
     }
+    if (!plan.files.some((file) => isJavaScriptFile(file.path))) {
+      throw new Error('Static build validation requires at least one checkable JavaScript source file.')
+    }
   }
 
   if (hasCommand(plan.commands, 'npm run lint')) {

From cd1fa8af74ac1e7d6a2c80135fd156c2aeebe413 Mon Sep 17 00:00:00 2001
From: Joncallim <64296013+Joncallim@users.noreply.github.com>
Date: Fri, 10 Jul 2026 15:58:16 +0800
Subject: [PATCH 7/9] Handle regex literals in focused test validation

---
 docs/acp-zed-connector.md                   |  7 +--
 web/__tests__/work-package-executor.test.ts |  2 +
 web/worker/work-package-executor.ts         | 49 ++++++++++++++++++++-
 3 files changed, 54 insertions(+), 4 deletions(-)

diff --git a/docs/acp-zed-connector.md b/docs/acp-zed-connector.md
index 960610e..5985238 100644
--- a/docs/acp-zed-connector.md
+++ b/docs/acp-zed-connector.md
@@ -83,9 +83,10 @@ acceptable.
 - Forge does not receive detailed token usage from ACP.
 - Tool calls from the underlying coding agent are not exposed as Forge runtime
   MCP grants.
-- ACP package execution is separately opt-in. When enabled, it relies on the
-  package attempt working directory and Forge's execution JSON path guards; it is
-  not an OS sandbox or a general live tool grant.
+- ACP package execution is enabled after task approval unless the operator sets
+  `FORGE_ACP_WORK_PACKAGE_EXECUTION=0`. When active, it relies on the package
+  attempt working directory and Forge's execution JSON path guards; it is not an
+  OS sandbox or a general live tool grant.
 - If a runtime does not expose model selection through ACP, Forge stores the
   selected model on the provider record but cannot force the local runtime to
   use it.
diff --git a/web/__tests__/work-package-executor.test.ts b/web/__tests__/work-package-executor.test.ts
index 429b8e1..350cf7e 100644
--- a/web/__tests__/work-package-executor.test.ts
+++ b/web/__tests__/work-package-executor.test.ts
@@ -1744,9 +1744,11 @@ describe('executeWorkPackage', () => {
             content: [
               'const test = require("node:test");',
               'const assert = require("node:assert/strict");',
+              "const quotePattern = /['\"]/;",
               'test("persists with a localStorage stub", () => {',
               '  const localStorageStub = new Map([["tasks", "[]"]]);',
               '  const input = { placeholder: "Add task" };',
+              '  assert.equal(quotePattern.test("\\\""), true);',
               '  assert.equal(localStorageStub.get("tasks"), "[]");',
               '  assert.equal(input.placeholder, "Add task");',
               '});',
diff --git a/web/worker/work-package-executor.ts b/web/worker/work-package-executor.ts
index a75735a..8252b9b 100644
--- a/web/worker/work-package-executor.ts
+++ b/web/worker/work-package-executor.ts
@@ -484,8 +484,10 @@ function isNoOpScript(script: string): boolean {
 
 function scannedJavaScriptText(content: string, stripStrings: boolean): string {
   const output = content.split('')
-  let state: 'code' | 'single' | 'double' | 'template' | 'line-comment' | 'block-comment' = 'code'
+  let state: 'code' | 'single' | 'double' | 'template' | 'regex' | 'line-comment' | 'block-comment' = 'code'
   let escaped = false
+  let regexAllowed = true
+  let regexCharacterClass = false
 
   const blank = (index: number) => {
     if (output[index] !== '\n' && output[index] !== '\r') output[index] = ' '
@@ -508,13 +510,37 @@ function scannedJavaScriptText(content: string, stripStrings: boolean): string {
         state = 'block-comment'
       } else if (char === "'") {
         if (stripStrings) blank(index)
+        escaped = false
         state = 'single'
       } else if (char === '"') {
         if (stripStrings) blank(index)
+        escaped = false
         state = 'double'
       } else if (char === '`') {
         if (stripStrings) blank(index)
+        escaped = false
         state = 'template'
+      } else if (char === '/' && regexAllowed) {
+        blank(index)
+        escaped = false
+        regexCharacterClass = false
+        state = 'regex'
+      } else if (/\s/.test(char)) {
+        continue
+      } else if (/[A-Za-z_$]/.test(char)) {
+        let end = index + 1
+        while (end < content.length && /[\w$]/.test(content[end])) end += 1
+        const word = content.slice(index, end)
+        regexAllowed = /^(?:await|case|delete|else|in|instanceof|of|return|throw|typeof|void|yield)$/.test(word)
+        index = end - 1
+      } else if (/\d/.test(char)) {
+        regexAllowed = false
+      } else if (char === ')' || char === ']' || char === '}') {
+        regexAllowed = false
+      } else if (char === '.' || (char === '+' && next === '+') || (char === '-' && next === '-')) {
+        regexAllowed = false
+      } else {
+        regexAllowed = true
       }
       continue
     }
@@ -537,6 +563,26 @@ function scannedJavaScriptText(content: string, stripStrings: boolean): string {
       continue
     }
 
+    if (state === 'regex') {
+      blank(index)
+      if (char === '\n' || char === '\r') {
+        state = 'code'
+        regexAllowed = true
+      } else if (escaped) {
+        escaped = false
+      } else if (char === '\\') {
+        escaped = true
+      } else if (char === '[') {
+        regexCharacterClass = true
+      } else if (char === ']') {
+        regexCharacterClass = false
+      } else if (char === '/' && !regexCharacterClass) {
+        state = 'code'
+        regexAllowed = false
+      }
+      continue
+    }
+
     if (stripStrings) blank(index)
     if (escaped) {
       escaped = false
@@ -548,6 +594,7 @@ function scannedJavaScriptText(content: string, stripStrings: boolean): string {
       (state === 'template' && char === '`')
     ) {
       state = 'code'
+      regexAllowed = false
     }
   }
 

From 4fec15604286fb0a53c186539080c17cc9751da6 Mon Sep 17 00:00:00 2001
From: Joncallim <64296013+Joncallim@users.noreply.github.com>
Date: Fri, 10 Jul 2026 16:08:33 +0800
Subject: [PATCH 8/9] Require assertions inside registered tests

---
 web/__tests__/work-package-executor.test.ts |  2 +-
 web/worker/work-package-executor.ts         | 63 ++++++++++++++++++++-
 2 files changed, 61 insertions(+), 4 deletions(-)

diff --git a/web/__tests__/work-package-executor.test.ts b/web/__tests__/work-package-executor.test.ts
index 350cf7e..769d81b 100644
--- a/web/__tests__/work-package-executor.test.ts
+++ b/web/__tests__/work-package-executor.test.ts
@@ -2019,7 +2019,7 @@ describe('executeWorkPackage', () => {
             },
             {
               path: 'tracker.test.js',
-              content: 'const test = require("node:test"); const assert = require("node:assert/strict"); test("adds item", () => console.log("ok"));\n',
+              content: 'const test = require("node:test"); const assert = require("node:assert/strict"); assert.equal(1, 1); test("adds item", () => {});\n',
             },
             {
               path: 'build-check.js',
diff --git a/web/worker/work-package-executor.ts b/web/worker/work-package-executor.ts
index 8252b9b..a88d15d 100644
--- a/web/worker/work-package-executor.ts
+++ b/web/worker/work-package-executor.ts
@@ -609,6 +609,65 @@ function hasExecutableModuleReference(commentFree: string, executable: string, p
   })
 }
 
+function findClosingParenthesis(content: string, openingIndex: number): number {
+  let depth = 0
+  for (let index = openingIndex; index < content.length; index += 1) {
+    if (content[index] === '(') depth += 1
+    if (content[index] !== ')') continue
+    depth -= 1
+    if (depth === 0) return index
+  }
+  return -1
+}
+
+function splitTopLevelArguments(content: string): string[] {
+  const argumentsList: string[] = []
+  let parentheses = 0
+  let brackets = 0
+  let braces = 0
+  let start = 0
+
+  for (let index = 0; index < content.length; index += 1) {
+    const char = content[index]
+    if (char === '(') parentheses += 1
+    else if (char === ')') parentheses -= 1
+    else if (char === '[') brackets += 1
+    else if (char === ']') brackets -= 1
+    else if (char === '{') braces += 1
+    else if (char === '}') braces -= 1
+    else if (char === ',' && parentheses === 0 && brackets === 0 && braces === 0) {
+      argumentsList.push(content.slice(start, index))
+      start = index + 1
+    }
+  }
+  argumentsList.push(content.slice(start))
+  return argumentsList
+}
+
+function hasAssertionInRegisteredTest(executable: string): boolean {
+  const assertionPattern = /\b(?:assert(?:\.[A-Za-z_$][\w$]*)?|ok|equal|strictEqual|deepEqual|deepStrictEqual|throws|rejects)\s*\(/
+  for (const match of executable.matchAll(/\b(?:test|it)(?:\.only)?\s*\(/g)) {
+    const openingIndex = (match.index ?? 0) + match[0].lastIndexOf('(')
+    const closingIndex = findClosingParenthesis(executable, openingIndex)
+    if (closingIndex < 0) continue
+
+    const callArguments = splitTopLevelArguments(executable.slice(openingIndex + 1, closingIndex))
+      .filter((argument) => argument.trim() !== '')
+    const callback = callArguments.at(-1) ?? ''
+    const arrowIndex = callback.indexOf('=>')
+    const functionIndex = callback.search(/\bfunction\b/)
+    const callbackBodyIndex = [
+      arrowIndex >= 0 ? arrowIndex + 2 : -1,
+      functionIndex >= 0 ? functionIndex + 'function'.length : -1,
+    ].filter((index) => index >= 0).sort((left, right) => left - right)[0]
+
+    if (callbackBodyIndex !== undefined && assertionPattern.test(callback.slice(callbackBodyIndex))) {
+      return true
+    }
+  }
+  return false
+}
+
 function isFocusedNodeTestAssertion(content: string): boolean {
   const commentFree = scannedJavaScriptText(content, false)
   const executable = scannedJavaScriptText(content, true)
@@ -622,9 +681,7 @@ function isFocusedNodeTestAssertion(content: string): boolean {
     executable,
     /\brequire\s*\(\s*['"]node:assert\/strict['"]\s*\)|\bfrom\s*['"]node:assert\/strict['"]|\bimport\s*['"]node:assert\/strict['"]/g,
   )
-  const hasTestCall = /\b(?:test|it)(?:\.only)?\s*\(/.test(executable)
-  const hasAssertionCall = /\b(?:assert(?:\.[A-Za-z_$][\w$]*)?|ok|equal|strictEqual|deepEqual|deepStrictEqual|throws|rejects)\s*\(/.test(executable)
-  return hasNodeTestImport && hasNodeAssertImport && hasTestCall && hasAssertionCall
+  return hasNodeTestImport && hasNodeAssertImport && hasAssertionInRegisteredTest(executable)
 }
 
 function isInvalidNodeTestScript(script: string): boolean {

From bfa2c407148522b00ba8e3a58231e109569aa1dd Mon Sep 17 00:00:00 2001
From: Joncallim <64296013+Joncallim@users.noreply.github.com>
Date: Fri, 10 Jul 2026 16:16:00 +0800
Subject: [PATCH 9/9] Reject custom runner test calls

---
 web/__tests__/work-package-executor.test.ts | 10 +++++++++-
 web/worker/work-package-executor.ts         |  2 +-
 2 files changed, 10 insertions(+), 2 deletions(-)

diff --git a/web/__tests__/work-package-executor.test.ts b/web/__tests__/work-package-executor.test.ts
index 769d81b..71940b5 100644
--- a/web/__tests__/work-package-executor.test.ts
+++ b/web/__tests__/work-package-executor.test.ts
@@ -2019,7 +2019,15 @@ describe('executeWorkPackage', () => {
             },
             {
               path: 'tracker.test.js',
-              content: 'const test = require("node:test"); const assert = require("node:assert/strict"); assert.equal(1, 1); test("adds item", () => {});\n',
+              content: [
+                'const test = require("node:test");',
+                'const assert = require("node:assert/strict");',
+                'const runner = { test: (callback) => callback() };',
+                'assert.equal(1, 1);',
+                'runner.test(() => assert.equal(1, 1));',
+                'test("adds item", () => {});',
+                '',
+              ].join('\n'),
             },
             {
               path: 'build-check.js',
diff --git a/web/worker/work-package-executor.ts b/web/worker/work-package-executor.ts
index a88d15d..f08b3d8 100644
--- a/web/worker/work-package-executor.ts
+++ b/web/worker/work-package-executor.ts
@@ -646,7 +646,7 @@ function splitTopLevelArguments(content: string): string[] {
 
 function hasAssertionInRegisteredTest(executable: string): boolean {
   const assertionPattern = /\b(?:assert(?:\.[A-Za-z_$][\w$]*)?|ok|equal|strictEqual|deepEqual|deepStrictEqual|throws|rejects)\s*\(/
-  for (const match of executable.matchAll(/\b(?:test|it)(?:\.only)?\s*\(/g)) {
+  for (const match of executable.matchAll(/(?:^|[^\w$.])(?:test|it)(?:\.only)?\s*\(/g)) {
     const openingIndex = (match.index ?? 0) + match[0].lastIndexOf('(')
     const closingIndex = findClosingParenthesis(executable, openingIndex)
     if (closingIndex < 0) continue