contextos-agents 1.6.1 → 2.0.0-beta.2

package/benchmarks/v2/pilot-tasks.json

@@ -0,0 +1,392 @@
+[
+  {
+    "id": "task-01-mock-task",
+    "description": "This is mock task 1 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-01.js"
+      ],
+      "filesToContain": {
+        "output-01.js": "console.log('mock 1');"
+      }
+    },
+    "hash": "sha256:7afaa346b4bf92bf9dc21e9ae809887412a86beb766842e99df7fee6573a4781"
+  },
+  {
+    "id": "task-02-mock-task",
+    "description": "This is mock task 2 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-02.js"
+      ],
+      "filesToContain": {
+        "output-02.js": "console.log('mock 2');"
+      }
+    },
+    "hash": "sha256:78686d2704d4d6900bb73f4941aa661a26a09f4346ab5e83281c0a18830ad1dd"
+  },
+  {
+    "id": "task-03-mock-task",
+    "description": "This is mock task 3 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-03.js"
+      ],
+      "filesToContain": {
+        "output-03.js": "console.log('mock 3');"
+      }
+    },
+    "hash": "sha256:1724980aa084bbaa16a8a69b664b783799bc75f8abe13ffc65b76843564fbbff"
+  },
+  {
+    "id": "task-04-mock-task",
+    "description": "This is mock task 4 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-04.js"
+      ],
+      "filesToContain": {
+        "output-04.js": "console.log('mock 4');"
+      }
+    },
+    "hash": "sha256:8f9ba36d8e59dd863b941cf6aba8709674901dba17f3e9c904e4da733538e38a"
+  },
+  {
+    "id": "task-05-mock-task",
+    "description": "This is mock task 5 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-05.js"
+      ],
+      "filesToContain": {
+        "output-05.js": "console.log('mock 5');"
+      }
+    },
+    "hash": "sha256:e7c08de777760acda054e9e20c19133287d1c245ef5efb6abaea6ec1ee3020e9"
+  },
+  {
+    "id": "task-06-mock-task",
+    "description": "This is mock task 6 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-06.js"
+      ],
+      "filesToContain": {
+        "output-06.js": "console.log('mock 6');"
+      }
+    },
+    "hash": "sha256:98c4ca9fa7207111d5816fe81f22b0e0cd27dc2b9d78ee9deb626d0a901f6921"
+  },
+  {
+    "id": "task-07-mock-task",
+    "description": "This is mock task 7 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-07.js"
+      ],
+      "filesToContain": {
+        "output-07.js": "console.log('mock 7');"
+      }
+    },
+    "hash": "sha256:014e4ceaa5a9e5112150d2536f78ecab3801305b5e95ceabf1de359314f9274f"
+  },
+  {
+    "id": "task-08-mock-task",
+    "description": "This is mock task 8 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-08.js"
+      ],
+      "filesToContain": {
+        "output-08.js": "console.log('mock 8');"
+      }
+    },
+    "hash": "sha256:f0470f4ad3dcb981ec0447de20036801ec638a0b7b32ae4dd8e6d1c94c08257e"
+  },
+  {
+    "id": "task-09-mock-task",
+    "description": "This is mock task 9 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-09.js"
+      ],
+      "filesToContain": {
+        "output-09.js": "console.log('mock 9');"
+      }
+    },
+    "hash": "sha256:e34d3152baeb112c9bd77f2381ea0cf9b902613c95d5ae2660b51a4e7ede2785"
+  },
+  {
+    "id": "task-10-mock-task",
+    "description": "This is mock task 10 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-10.js"
+      ],
+      "filesToContain": {
+        "output-10.js": "console.log('mock 10');"
+      }
+    },
+    "hash": "sha256:8bfdd14ef40300544ba24d517506314c6b8fd3075ecad82341f15eb2715692e0"
+  },
+  {
+    "id": "task-11-mock-task",
+    "description": "This is mock task 11 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-11.js"
+      ],
+      "filesToContain": {
+        "output-11.js": "console.log('mock 11');"
+      }
+    },
+    "hash": "sha256:555ada841ff8a3cc6b4ce15165f2d214f818bde82cdb71d942137ab132d499e6"
+  },
+  {
+    "id": "task-12-mock-task",
+    "description": "This is mock task 12 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-12.js"
+      ],
+      "filesToContain": {
+        "output-12.js": "console.log('mock 12');"
+      }
+    },
+    "hash": "sha256:712690d9849664fd0adbfeb295d9e47618024ad8c4be52adfa314e3d31481e50"
+  },
+  {
+    "id": "task-13-mock-task",
+    "description": "This is mock task 13 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-13.js"
+      ],
+      "filesToContain": {
+        "output-13.js": "console.log('mock 13');"
+      }
+    },
+    "hash": "sha256:aa050d240a6c17ad68f304685f77782bbb8ab5cc6dae23d3cbd845de6cf87304"
+  },
+  {
+    "id": "task-14-mock-task",
+    "description": "This is mock task 14 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-14.js"
+      ],
+      "filesToContain": {
+        "output-14.js": "console.log('mock 14');"
+      }
+    },
+    "hash": "sha256:ec88ecb7025218c79945d70e568f358e61452ab70d9c434b337411f6a8f468fc"
+  },
+  {
+    "id": "task-15-mock-task",
+    "description": "This is mock task 15 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-15.js"
+      ],
+      "filesToContain": {
+        "output-15.js": "console.log('mock 15');"
+      }
+    },
+    "hash": "sha256:5f1f6caf37ba31b3b91b876e5e592717abc9c97e0349c15e7ff3c2da2d133947"
+  },
+  {
+    "id": "task-16-mock-task",
+    "description": "This is mock task 16 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-16.js"
+      ],
+      "filesToContain": {
+        "output-16.js": "console.log('mock 16');"
+      }
+    },
+    "hash": "sha256:e8e6bfe60cd69b3373e667336d41d3c95fdbc5008f9c0eb99b1b94460e7ad0fd"
+  },
+  {
+    "id": "task-17-mock-task",
+    "description": "This is mock task 17 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-17.js"
+      ],
+      "filesToContain": {
+        "output-17.js": "console.log('mock 17');"
+      }
+    },
+    "hash": "sha256:3a41c6b074a0770e0fb6822ed252c69a77c05fa7bd97fcd9bf9b16f6cf193dd8"
+  },
+  {
+    "id": "task-18-mock-task",
+    "description": "This is mock task 18 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-18.js"
+      ],
+      "filesToContain": {
+        "output-18.js": "console.log('mock 18');"
+      }
+    },
+    "hash": "sha256:dbd6cc88cb73c805eef95f0cea09e09927f6b13e33fad1e69c049d174eab537f"
+  },
+  {
+    "id": "task-19-mock-task",
+    "description": "This is mock task 19 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-19.js"
+      ],
+      "filesToContain": {
+        "output-19.js": "console.log('mock 19');"
+      }
+    },
+    "hash": "sha256:b844124ce8543a3863a82699dd4cefb5e9b9391547934546974419be2b069895"
+  },
+  {
+    "id": "task-20-mock-task",
+    "description": "This is mock task 20 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-20.js"
+      ],
+      "filesToContain": {
+        "output-20.js": "console.log('mock 20');"
+      }
+    },
+    "hash": "sha256:ad1eb8063f7138e9fb193982a6bc4db5eed1e8fb4af21ddbda75270cb2c06d51"
+  },
+  {
+    "id": "task-21-mock-task",
+    "description": "This is mock task 21 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-21.js"
+      ],
+      "filesToContain": {
+        "output-21.js": "console.log('mock 21');"
+      }
+    },
+    "hash": "sha256:2dd74021abc2bb10e9bed627b7fb072311ed8e86b95526b8a57149cd4f89a7df"
+  },
+  {
+    "id": "task-22-mock-task",
+    "description": "This is mock task 22 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-22.js"
+      ],
+      "filesToContain": {
+        "output-22.js": "console.log('mock 22');"
+      }
+    },
+    "hash": "sha256:4865211d3986fbea264350c8965dba461d6eb5dec27beceaebd247c49b7f6fff"
+  },
+  {
+    "id": "task-23-mock-task",
+    "description": "This is mock task 23 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-23.js"
+      ],
+      "filesToContain": {
+        "output-23.js": "console.log('mock 23');"
+      }
+    },
+    "hash": "sha256:168e4241ee5aee2f186f0aab6856178a11fef6399d27f17f62cc25cb695aac0e"
+  },
+  {
+    "id": "task-24-mock-task",
+    "description": "This is mock task 24 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-24.js"
+      ],
+      "filesToContain": {
+        "output-24.js": "console.log('mock 24');"
+      }
+    },
+    "hash": "sha256:1b7f1e388d8f3e020f6f33a61f29bf40f9e56f203f1b634c47b36e70a5493f01"
+  },
+  {
+    "id": "task-25-mock-task",
+    "description": "This is mock task 25 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-25.js"
+      ],
+      "filesToContain": {
+        "output-25.js": "console.log('mock 25');"
+      }
+    },
+    "hash": "sha256:2ad0d0c64bbc6d5fc6bb10e398ce1e62ae9a6c73b4a13390646d94f90830ee04"
+  },
+  {
+    "id": "task-26-mock-task",
+    "description": "This is mock task 26 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-26.js"
+      ],
+      "filesToContain": {
+        "output-26.js": "console.log('mock 26');"
+      }
+    },
+    "hash": "sha256:b4eaa7cfe62d917b87a0311079eadd9a7c2488676f84b3bad540425c5d52ed5c"
+  },
+  {
+    "id": "task-27-mock-task",
+    "description": "This is mock task 27 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-27.js"
+      ],
+      "filesToContain": {
+        "output-27.js": "console.log('mock 27');"
+      }
+    },
+    "hash": "sha256:1d571536957594de46fbd9fe0710eefcc66d849ea53dde9ad61ee483941349e1"
+  },
+  {
+    "id": "task-28-mock-task",
+    "description": "This is mock task 28 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-28.js"
+      ],
+      "filesToContain": {
+        "output-28.js": "console.log('mock 28');"
+      }
+    },
+    "hash": "sha256:765672c715bfb9813a05949c7e9eff2a8cd91eed05b0ee026d177561b77411eb"
+  },
+  {
+    "id": "task-29-mock-task",
+    "description": "This is mock task 29 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-29.js"
+      ],
+      "filesToContain": {
+        "output-29.js": "console.log('mock 29');"
+      }
+    },
+    "hash": "sha256:7209b936824e13b1ec6eeec1503b0ef2ffd0698378ab5739857130773ebf743c"
+  },
+  {
+    "id": "task-30-mock-task",
+    "description": "This is mock task 30 for the benchmark pilot.",
+    "expectedState": {
+      "filesToExist": [
+        "output-30.js"
+      ],
+      "filesToContain": {
+        "output-30.js": "console.log('mock 30');"
+      }
+    },
+    "hash": "sha256:872fd6ef064d17a4315293952bf4d8704c47d3d94affa903695dfd7e2f9461e8"
+  }
+]

package/bin/commands/recover.js

@@ -0,0 +1,88 @@
+'use strict';
+
+const path = require('path');
+const { JournaledTransaction, TX_STATES } = require('../../.agents/filesystem/journaled-transaction.js');
+const { ProjectMutationLock } = require('../../.agents/filesystem/project-lock.js');
+
+// Minimal ANSI colors
+const isTTY = process.stdout.isTTY;
+const c = {
+  red: (s) => isTTY ? `\x1b[31m${s}\x1b[0m` : s,
+  green: (s) => isTTY ? `\x1b[32m${s}\x1b[0m` : s,
+  yellow: (s) => isTTY ? `\x1b[33m${s}\x1b[0m` : s,
+  cyan: (s) => isTTY ? `\x1b[36m${s}\x1b[0m` : s,
+};
+
+function recoverCommand(args, flags) {
+  const ROOT = process.cwd();
+
+  const isList = flags.list || args.includes('--list');
+  
+  const rollbackIdx = args.indexOf('--rollback');
+  const txIdRollback = rollbackIdx !== -1 && args[rollbackIdx + 1] ? args[rollbackIdx + 1] : null;
+
+  const resumeIdx = args.indexOf('--resume');
+  const txIdResume = resumeIdx !== -1 && args[resumeIdx + 1] ? args[resumeIdx + 1] : null;
+
+  if (isList) {
+    const pending = JournaledTransaction.listPending(ROOT);
+    if (pending.length === 0) {
+      console.log(c.green('✓ No pending or stuck transactions found.'));
+      process.exit(0);
+    }
+    console.log(c.yellow(`Found ${pending.length} pending/stuck transaction(s):\n`));
+    pending.forEach(tx => {
+      console.log(`  ID:    ${c.cyan(tx.txId)}`);
+      console.log(`  State: ${tx.state === TX_STATES.RECOVERY_REQUIRED ? c.red(tx.state) : c.yellow(tx.state)}`);
+      console.log(`  Date:  ${tx.createdAt}`);
+      if (tx.error) console.log(`  Error: ${c.red(tx.error)}`);
+      console.log('');
+    });
+    console.log('To recover, run:');
+    console.log(`  contextos recover --resume <ID>`);
+    console.log(`  contextos recover --rollback <ID>`);
+    process.exit(0);
+  }
+
+  if (txIdRollback) {
+    console.log(c.cyan(`Attempting to rollback transaction: ${txIdRollback}`));
+    const projectLock = new ProjectMutationLock(ROOT);
+    let lockToken = null;
+    try {
+      lockToken = projectLock.acquire({ command: 'recover:rollback' });
+      const tx = JournaledTransaction.load(ROOT, txIdRollback);
+      tx.rollback();
+      console.log(c.green(`✓ Transaction ${txIdRollback} successfully rolled back.`));
+    } catch (err) {
+      console.error(c.red(`\n[ERROR] Rollback failed: ${err.message}`));
+      process.exit(1);
+    } finally {
+      if (lockToken) projectLock.release(lockToken);
+    }
+    process.exit(0);
+  }
+
+  if (txIdResume) {
+    console.log(c.cyan(`Attempting to resume transaction: ${txIdResume}`));
+    const projectLock = new ProjectMutationLock(ROOT);
+    let lockToken = null;
+    try {
+      lockToken = projectLock.acquire({ command: 'recover:resume' });
+      const tx = JournaledTransaction.load(ROOT, txIdResume);
+      // Simply calling commit() will resume if it's not COMMITTED/ROLLED_BACK
+      tx.commit();
+      console.log(c.green(`✓ Transaction ${txIdResume} successfully resumed and committed.`));
+    } catch (err) {
+      console.error(c.red(`\n[ERROR] Resume failed: ${err.message}`));
+      process.exit(1);
+    } finally {
+      if (lockToken) projectLock.release(lockToken);
+    }
+    process.exit(0);
+  }
+
+  console.log(c.red('Usage: contextos recover [--list] [--rollback <txId>] [--resume <txId>]'));
+  process.exit(1);
+}
+
+module.exports = recoverCommand;

package/bin/commands/uninstall.js

@@ -0,0 +1,207 @@
+/**
+ * bin/commands/uninstall.js
+ * ContextOS CLI — Safe Uninstall Command
+ *
+ * Removes ContextOS artifacts safely:
+ * - Only deletes clean managed files (content matches installation hash)
+ * - Never deletes user-modified managed files (USER_MODIFIED_PRESERVED)
+ * - Never deletes user-created custom skills or unmanaged files
+ * - Preserves .agents/ directory if user files remain
+ * - Removes .agents/ cleanly if no user files remain
+ * - Supports --dry-run
+ */
+
+const fs = require('fs');
+const path = require('path');
+const {
+  loadLockfile,
+  migrateExistingInstallation,
+} = require('../lib/lockfile.js');
+const {
+  planFileDelete,
+  executeFileDelete,
+} = require('../lib/safe-writer.js');
+
+/**
+ * Recursively count all files remaining in a directory (excluding specified skip names).
+ */
+function countFiles(dir, skipNames = new Set()) {
+  let count = 0;
+  if (!fs.existsSync(dir)) return 0;
+
+  try {
+    for (const entry of fs.readdirSync(dir, { withFileTypes: true })) {
+      if (skipNames.has(entry.name)) continue;
+      const full = path.join(dir, entry.name);
+      if (entry.isDirectory()) {
+        count += countFiles(full, skipNames);
+      } else if (entry.isFile()) {
+        count++;
+      }
+    }
+  } catch {
+    return 0;
+  }
+  return count;
+}
+
+/**
+ * Plan uninstallation for a target project.
+ */
+function planUninstall(projectDir, options = {}) {
+  const targetAgentsDir = path.join(projectDir, '.agents');
+  if (!fs.existsSync(targetAgentsDir)) {
+    return {
+      ok: false,
+      error: 'NO_AGENTS_DIR',
+      message: 'No .agents/ directory found in the target project. Nothing to uninstall.',
+      plans: [],
+      summary: { delete: 0, skipModified: 0, skipUnmanaged: 0, missing: 0 },
+    };
+  }
+
+  const sourceAgentsDir = options.sourceAgentsDir || path.join(__dirname, '..', '..', '.agents');
+
+  // Load existing lockfile or build migration plan
+  let lockData = loadLockfile(projectDir);
+  if (!lockData) {
+    lockData = migrateExistingInstallation(projectDir, sourceAgentsDir, {
+      dryRun: true,
+      inMemory: true,
+    });
+  }
+
+  const plans = [];
+  if (lockData && lockData.managedFiles) {
+    for (const relPath of Object.keys(lockData.managedFiles)) {
+      const plan = planFileDelete({
+        projectDir,
+        relPath,
+        lockData,
+      });
+      plans.push(plan);
+    }
+  }
+
+  // Count unmanaged files in .agents that are not in lockData
+  const managedSet = new Set(Object.keys(lockData?.managedFiles || {}));
+  let unmanagedFilesCount = 0;
+
+  function findUnmanaged(currentDir) {
+    let entries;
+    try {
+      entries = fs.readdirSync(currentDir, { withFileTypes: true });
+    } catch {
+      return;
+    }
+
+    for (const entry of entries) {
+      const full = path.join(currentDir, entry.name);
+      if (entry.isDirectory()) {
+        findUnmanaged(full);
+      } else if (entry.isFile()) {
+        if (entry.name === 'contextos.lock.json') continue;
+        const rel = path.relative(projectDir, full).replace(/\\/g, '/');
+        if (!managedSet.has(rel)) {
+          unmanagedFilesCount++;
+        }
+      }
+    }
+  }
+
+  findUnmanaged(targetAgentsDir);
+
+  const summary = {
+    delete: plans.filter(p => p.action === 'DELETE').length,
+    skipModified: plans.filter(p => p.action === 'SKIP' && p.reason === 'USER_MODIFIED_PRESERVED').length,
+    missing: plans.filter(p => p.action === 'SKIP' && p.reason === 'ALREADY_MISSING').length,
+    skipUnmanaged: unmanagedFilesCount,
+  };
+
+  return {
+    ok: true,
+    plans,
+    summary,
+    lockData,
+  };
+}
+
+/**
+ * Run safe uninstall against a target project.
+ */
+function runUninstall(projectDir = process.cwd(), options = {}) {
+  const dryRun = Boolean(options.dryRun);
+
+  console.log(`\nContextOS Safe Uninstall ${dryRun ? '[DRY-RUN]' : ''}`);
+  console.log(`Target: ${projectDir}`);
+
+  const planned = planUninstall(projectDir, options);
+  if (!planned.ok) {
+    console.error(`\n[ERROR] ${planned.message}`);
+    return planned;
+  }
+
+  const { lockData, plans, summary } = planned;
+
+  console.log('\nPlanned Actions:');
+  for (const item of plans) {
+    if (item.action === 'DELETE') {
+      console.log(`  - DELETE    ${item.relPath}`);
+    } else if (item.reason === 'USER_MODIFIED_PRESERVED') {
+      console.log(`  = PRESERVE  ${item.relPath} (modified by user — will NOT delete)`);
+    }
+  }
+
+  if (summary.skipUnmanaged > 0) {
+    console.log(`  = PRESERVE  ${summary.skipUnmanaged} unmanaged user file(s)/skill(s) in .agents/`);
+  }
+
+  console.log('\nSummary:');
+  console.log(`  To delete:     ${summary.delete}`);
+  console.log(`  Preserved:     ${summary.skipModified} (user-modified) + ${summary.skipUnmanaged} (user-created)`);
+
+  if (dryRun) {
+    console.log('\n[DRY-RUN] No files were deleted.');
+    return { ...planned, executed: false };
+  }
+
+  // Execute deletions
+  let deletedCount = 0;
+  for (const item of plans) {
+    if (item.action === 'DELETE') {
+      const res = executeFileDelete(item, lockData, { dryRun: false });
+      if (res.executed) deletedCount++;
+    }
+  }
+
+  const targetAgentsDir = path.join(projectDir, '.agents');
+  const lockfilePath = path.join(targetAgentsDir, 'contextos.lock.json');
+
+  // Count remaining files in .agents (excluding lockfile)
+  const remainingFiles = countFiles(targetAgentsDir, new Set(['contextos.lock.json']));
+
+  if (remainingFiles === 0) {
+    // Completely clean, remove lockfile and .agents directory
+    try {
+      if (fs.existsSync(lockfilePath)) fs.unlinkSync(lockfilePath);
+      fs.rmSync(targetAgentsDir, { recursive: true, force: true });
+      console.log('\n[OK] ContextOS completely uninstalled. .agents/ removed.');
+    } catch (err) {
+      console.warn(`[WARN] Could not remove .agents directory: ${err.message}`);
+    }
+  } else {
+    // User files or modified files remain — remove lockfile or keep remaining?
+    try {
+      if (fs.existsSync(lockfilePath)) fs.unlinkSync(lockfilePath);
+    } catch {}
+    console.log(`\n[OK] Clean ContextOS artifacts uninstalled (${deletedCount} removed).`);
+    console.log(`[INFO] .agents/ preserved because ${remainingFiles} user-modified or custom file(s) remain.`);
+  }
+
+  return { ...planned, executed: true, deletedCount, remainingFiles };
+}
+
+module.exports = {
+  planUninstall,
+  runUninstall,
+};