mstro-app 0.3.0 → 0.3.4

package/dist/server/mcp/bouncer-cli.js

@@ -1,99 +0,0 @@
-#!/usr/bin/env node
-// Copyright (c) 2025-present Mstro, Inc. All rights reserved.
-// Licensed under the MIT License. See LICENSE file for details.
-/**
- * Bouncer CLI - Shell-callable wrapper for Mstro security bouncer
- *
- * This CLI reads Claude Code hook input from stdin and returns a security decision.
- * It's designed to be called from bouncer.sh.
- *
- * Input (stdin): Claude Code PreToolUse hook JSON payload
- * Output (stdout): JSON decision { decision: "allow"|"deny", reason: string }
- *
- * The hook payload includes conversation context that we pass to the bouncer
- * so it can make context-aware decisions.
- */
-import { reviewOperation } from './bouncer-integration.js';
-/**
- * Read all data from stdin (Node.js compatible)
- */
-async function readStdin() {
-    return new Promise((resolve, reject) => {
-        const chunks = [];
-        process.stdin.on('data', (chunk) => chunks.push(chunk));
-        process.stdin.on('end', () => resolve(Buffer.concat(chunks).toString('utf-8').trim()));
-        process.stdin.on('error', reject);
-    });
-}
-function buildOperationString(toolName, toolInput) {
-    if (toolName === 'Bash' && toolInput.command) {
-        return `${toolName}: ${toolInput.command}`;
-    }
-    if (['Write', 'Edit', 'Read'].includes(toolName)) {
-        const filePath = toolInput.file_path || toolInput.filePath || toolInput.path;
-        return filePath ? `${toolName}: ${filePath}` : `${toolName}: ${JSON.stringify(toolInput)}`;
-    }
-    return `${toolName}: ${JSON.stringify(toolInput)}`;
-}
-function extractConversationContext(hookInput) {
-    const lastUserMessage = hookInput.conversation?.last_user_message;
-    if (lastUserMessage)
-        return `User's request: "${lastUserMessage}"`;
-    const recentMessages = hookInput.conversation?.messages?.slice(-5);
-    if (recentMessages?.length) {
-        return `Recent conversation:\n${recentMessages.map(m => `${m.role}: ${m.content}`).join('\n')}`;
-    }
-    return undefined;
-}
-async function main() {
-    const inputStr = await readStdin();
-    if (!inputStr) {
-        console.log(JSON.stringify({ decision: 'allow', reason: 'Empty input, allowing' }));
-        process.exit(0);
-    }
-    let hookInput;
-    try {
-        hookInput = JSON.parse(inputStr);
-    }
-    catch (e) {
-        console.error('[bouncer-cli] Failed to parse input JSON:', e);
-        console.log(JSON.stringify({ decision: 'allow', reason: 'Invalid JSON input, allowing' }));
-        process.exit(0);
-    }
-    const toolName = hookInput.tool_name || hookInput.toolName || 'unknown';
-    const toolInput = hookInput.input || hookInput.toolInput || {};
-    const userRequestContext = extractConversationContext(hookInput);
-    const lastUserMessage = hookInput.conversation?.last_user_message;
-    const recentMessages = hookInput.conversation?.messages?.slice(-5);
-    const bouncerRequest = {
-        operation: buildOperationString(toolName, toolInput),
-        context: {
-            purpose: userRequestContext || 'Tool use request from Claude',
-            workingDirectory: hookInput.working_directory || process.cwd(),
-            toolName,
-            toolInput,
-            userRequest: lastUserMessage,
-            conversationHistory: recentMessages?.map(m => `${m.role}: ${m.content}`),
-            sessionId: hookInput.session_id,
-        },
-    };
-    try {
-        const decision = await reviewOperation(bouncerRequest);
-        console.log(JSON.stringify({
-            decision: decision.decision === 'deny' ? 'deny' : 'allow',
-            reason: decision.reasoning,
-            confidence: decision.confidence,
-            threatLevel: decision.threatLevel,
-            alternative: decision.alternative,
-        }));
-    }
-    catch (error) {
-        console.error('[bouncer-cli] Error:', error.message);
-        console.log(JSON.stringify({
-            decision: 'allow',
-            reason: `Bouncer error: ${error.message}. Allowing to avoid blocking.`
-        }));
-    }
-}
-main();
-//# sourceMappingURL=bouncer-cli.js.map

package/dist/server/mcp/bouncer-cli.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"bouncer-cli.js","sourceRoot":"","sources":["../../../server/mcp/bouncer-cli.ts"],"names":[],"mappings":";AACA,8DAA8D;AAC9D,gEAAgE;AAEhE;;;;;;;;;;;GAWG;AAEH,OAAO,EAA6B,eAAe,EAAE,MAAM,0BAA0B,CAAC;AAqBtF;;GAEG;AACH,KAAK,UAAU,SAAS;IACtB,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACrC,MAAM,MAAM,GAAa,EAAE,CAAC;QAC5B,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;QACxD,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;QACvF,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;IACpC,CAAC,CAAC,CAAC;AACL,CAAC;AAED,SAAS,oBAAoB,CAAC,QAAgB,EAAE,SAA8B;IAC5E,IAAI,QAAQ,KAAK,MAAM,IAAI,SAAS,CAAC,OAAO,EAAE,CAAC;QAC7C,OAAO,GAAG,QAAQ,KAAK,SAAS,CAAC,OAAO,EAAE,CAAC;IAC7C,CAAC;IACD,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,EAAE,CAAC;QACjD,MAAM,QAAQ,GAAG,SAAS,CAAC,SAAS,IAAI,SAAS,CAAC,QAAQ,IAAI,SAAS,CAAC,IAAI,CAAC;QAC7E,OAAO,QAAQ,CAAC,CAAC,CAAC,GAAG,QAAQ,KAAK,QAAQ,EAAE,CAAC,CAAC,CAAC,GAAG,QAAQ,KAAK,IAAI,CAAC,SAAS,CAAC,SAAS,CAAC,EAAE,CAAC;IAC7F,CAAC;IACD,OAAO,GAAG,QAAQ,KAAK,IAAI,CAAC,SAAS,CAAC,SAAS,CAAC,EAAE,CAAC;AACrD,CAAC;AAED,SAAS,0BAA0B,CAAC,SAAoB;IACtD,MAAM,eAAe,GAAG,SAAS,CAAC,YAAY,EAAE,iBAAiB,CAAC;IAClE,IAAI,eAAe;QAAE,OAAO,oBAAoB,eAAe,GAAG,CAAC;IAEnE,MAAM,cAAc,GAAG,SAAS,CAAC,YAAY,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IACnE,IAAI,cAAc,EAAE,MAAM,EAAE,CAAC;QAC3B,OAAO,yBAAyB,cAAc,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;IAClG,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,KAAK,UAAU,IAAI;IACjB,MAAM,QAAQ,GAAG,MAAM,SAAS,EAAE,CAAC;IAEnC,IAAI,CAAC,QAAQ,EAAE,CAAC;QACd,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,uBAAuB,EAAE,CAAC,CAAC,CAAC;QACpF,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;IAED,IAAI,SAAoB,CAAC;IACzB,IAAI,CAAC;QACH,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;IACnC,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACX,OAAO,CAAC,KAAK,CAAC,2CAA2C,EAAE,CAAC,CAAC,CAAC;QAC9D,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,8BAA8B,EAAE,CAAC,CAAC,CAAC;QAC3F,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;IAED,MAAM,QAAQ,GAAG,SAAS,CAAC,SAAS,IAAI,SAAS,CAAC,QAAQ,IAAI,SAAS,CAAC;IACxE,MAAM,SAAS,GAAG,SAAS,CAAC,KAAK,IAAI,SAAS,CAAC,SAAS,IAAI,EAAE,CAAC;IAC/D,MAAM,kBAAkB,GAAG,0BAA0B,CAAC,SAAS,CAAC,CAAC;IACjE,MAAM,eAAe,GAAG,SAAS,CAAC,YAAY,EAAE,iBAAiB,CAAC;IAClE,MAAM,cAAc,GAAG,SAAS,CAAC,YAAY,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IAEnE,MAAM,cAAc,GAAyB;QAC3C,SAAS,EAAE,oBAAoB,CAAC,QAAQ,EAAE,SAAS,CAAC;QACpD,OAAO,EAAE;YACP,OAAO,EAAE,kBAAkB,IAAI,8BAA8B;YAC7D,gBAAgB,EAAE,SAAS,CAAC,iBAAiB,IAAI,OAAO,CAAC,GAAG,EAAE;YAC9D,QAAQ;YACR,SAAS;YACT,WAAW,EAAE,eAAe;YAC5B,mBAAmB,EAAE,cAAc,EAAE,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,OAAO,EAAE,CAAC;YACxE,SAAS,EAAE,SAAS,CAAC,UAAU;SAChC;KACF,CAAC;IAEF,IAAI,CAAC;QACH,MAAM,QAAQ,GAAG,MAAM,eAAe,CAAC,cAAc,CAAC,CAAC;QACvD,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC;YACzB,QAAQ,EAAE,QAAQ,CAAC,QAAQ,KAAK,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO;YACzD,MAAM,EAAE,QAAQ,CAAC,SAAS;YAC1B,UAAU,EAAE,QAAQ,CAAC,UAAU;YAC/B,WAAW,EAAE,QAAQ,CAAC,WAAW;YACjC,WAAW,EAAE,QAAQ,CAAC,WAAW;SAClC,CAAC,CAAC,CAAC;IACN,CAAC;IAAC,OAAO,KAAU,EAAE,CAAC;QACpB,OAAO,CAAC,KAAK,CAAC,sBAAsB,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;QACrD,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC;YACzB,QAAQ,EAAE,OAAO;YACjB,MAAM,EAAE,kBAAkB,KAAK,CAAC,OAAO,+BAA+B;SACvE,CAAC,CAAC,CAAC;IACN,CAAC;AACH,CAAC;AAED,IAAI,EAAE,CAAC"}

package/hooks/bouncer.sh

@@ -1,145 +0,0 @@
-#!/usr/bin/env bash
-# Copyright (c) 2025-present Mstro, Inc. All rights reserved.
-# Licensed under the MIT License. See LICENSE file for details.
-
-#
-# Mstro Bouncer Gate - Claude Code Hook
-#
-# This hook intercepts Claude Code tool calls and routes them through
-# the Mstro bouncer for security analysis before execution.
-#
-# Installation:
-#   Run: npx mstro --configure-hooks
-#
-# Dependencies: Node.js (no jq or bun required)
-#
-
-set -euo pipefail
-
-# Configuration - MSTRO_BOUNCER_CLI is set by configure-claude.js
-BOUNCER_CLI="${MSTRO_BOUNCER_CLI:-}"
-BOUNCER_TIMEOUT="${BOUNCER_TIMEOUT:-10}"
-BOUNCER_LOG="${BOUNCER_LOG:-$HOME/.claude/logs/bouncer.log}"
-
-# Ensure log directory exists
-mkdir -p "$(dirname "$BOUNCER_LOG")"
-
-# Read hook input from stdin (JSON format from Claude Code)
-INPUT=$(cat)
-
-# Use Node.js inline to parse JSON and handle logic (eliminates jq dependency)
-RESULT=$(node --input-type=module -e "
-import { spawn } from 'child_process';
-import { appendFileSync } from 'fs';
-
-const input = JSON.parse(process.argv[1]);
-const bouncerCli = process.argv[2];
-const timeout = parseInt(process.argv[3], 10) * 1000;
-const logFile = process.argv[4];
-
-const toolName = input.tool_name || input.toolName || 'unknown';
-const toolInput = input.input || input.toolInput || {};
-
-function log(msg) {
-  const timestamp = new Date().toISOString();
-  appendFileSync(logFile, \`[\${timestamp}] \${msg}\n\`);
-}
-
-function output(decision, reason) {
-  console.log(JSON.stringify({ decision, reason }));
-}
-
-// Quick path for read-only and side-effect-free operations
-const safeOps = ['Read', 'Glob', 'Grep', 'Search', 'List', 'WebFetch', 'WebSearch',
-  'ExitPlanMode', 'EnterPlanMode', 'TodoWrite', 'AskUserQuestion'];
-if (safeOps.includes(toolName)) {
-  output('allow', 'Safe operation (no dangerous side effects)');
-  process.exit(0);
-}
-
-// Quick path for malformed tool calls with empty params (no-ops)
-if (Object.keys(toolInput).length === 0 && ['Edit', 'Write'].includes(toolName)) {
-  output('allow', 'Empty parameters - no-op');
-  process.exit(0);
-}
-
-// Build operation string for logging
-let operation = toolName + ': ';
-if (toolName === 'Bash' && toolInput.command) {
-  operation += toolInput.command;
-} else if (['Write', 'Edit'].includes(toolName)) {
-  operation += toolInput.file_path || toolInput.filePath || toolInput.path || JSON.stringify(toolInput);
-} else {
-  operation += JSON.stringify(toolInput);
-}
-
-log('Analyzing: ' + toolName);
-
-// Check if bouncer CLI is available
-if (bouncerCli) {
-  try {
-    const child = spawn('node', [bouncerCli], {
-      stdio: ['pipe', 'pipe', 'pipe'],
-      timeout: timeout
-    });
-
-    let stdout = '';
-    let stderr = '';
-
-    child.stdout.on('data', (data) => { stdout += data; });
-    child.stderr.on('data', (data) => { stderr += data; });
-
-    child.stdin.write(JSON.stringify(input));
-    child.stdin.end();
-
-    child.on('close', (code) => {
-      if (stderr) log('Bouncer stderr: ' + stderr);
-
-      if (code === 0 && stdout.trim()) {
-        try {
-          const result = JSON.parse(stdout.trim());
-          log(result.decision + ': ' + operation + ' - ' + (result.reason || ''));
-          console.log(stdout.trim());
-        } catch {
-          log('allow (parse error): ' + operation);
-          output('allow', 'Bouncer response parse error, allowing');
-        }
-      } else {
-        log('allow (bouncer error): ' + operation);
-        output('allow', 'Bouncer error, allowing');
-      }
-    });
-
-    child.on('error', (err) => {
-      log('allow (spawn error): ' + operation + ' - ' + err.message);
-      output('allow', 'Bouncer spawn error, allowing');
-    });
-
-  } catch (err) {
-    log('allow (exception): ' + operation + ' - ' + err.message);
-    output('allow', 'Bouncer exception, allowing');
-  }
-} else {
-  // Fallback: critical threat pattern matching only
-  const criticalPatterns = [
-    { pattern: /rm\s+-rf\s+(\/|~)(\$|\s)/, reason: 'Critical threat: recursive delete of root or home' },
-    { pattern: /:\(\)\{.*\}|:\(\)\{.*:\|:/, reason: 'Critical threat: fork bomb detected' },
-    { pattern: /dd\s+if=\/dev\/zero\s+of=\/dev\/sd/, reason: 'Critical threat: disk overwrite' },
-    { pattern: /mkfs\s+\/dev\/sd/, reason: 'Critical threat: filesystem format' },
-    { pattern: />\s*\/dev\/sd/, reason: 'Critical threat: direct disk write' },
-  ];
-
-  for (const { pattern, reason } of criticalPatterns) {
-    if (pattern.test(operation)) {
-      log('DENIED: ' + operation + ' - ' + reason);
-      output('deny', reason);
-      process.exit(0);
-    }
-  }
-
-  log('allow (fallback): ' + operation);
-  output('allow', 'Basic pattern check passed');
-}
-" "$INPUT" "$BOUNCER_CLI" "$BOUNCER_TIMEOUT" "$BOUNCER_LOG" 2>> "$BOUNCER_LOG")
-
-echo "$RESULT"

package/server/cli/headless/output-utils.test.ts

@@ -1,225 +0,0 @@
-import { describe, expect, it } from 'vitest';
-import {
-  detectErrorInStderr,
-  estimateTokensFromOutput,
-  extractCleanOutput,
-  extractModifiedFiles,
-} from './output-utils.js';
-
-// ========== extractCleanOutput ==========
-
-describe('extractCleanOutput', () => {
-  it('filters out JSON lines with "type" field', () => {
-    const input = [
-      '{"type": "system", "data": "init"}',
-      'Hello world',
-      '{"type": "assistant", "text": "hi"}',
-      'Some output',
-    ].join('\n');
-    expect(extractCleanOutput(input)).toBe('Hello world\nSome output');
-  });
-
-  it('strips ANSI color codes', () => {
-    const input = '\x1b[32mgreen text\x1b[0m and \x1b[1;31mred bold\x1b[0m';
-    expect(extractCleanOutput(input)).toBe('green text and red bold');
-  });
-
-  it('normalizes CRLF to LF', () => {
-    const input = 'line1\r\nline2\r\nline3';
-    expect(extractCleanOutput(input)).toBe('line1\nline2\nline3');
-  });
-
-  it('trims whitespace', () => {
-    const input = '  \n  Hello  \n  ';
-    expect(extractCleanOutput(input)).toBe('Hello');
-  });
-
-  it('filters empty lines', () => {
-    const input = 'line1\n\n\nline2';
-    expect(extractCleanOutput(input)).toBe('line1\nline2');
-  });
-
-  it('returns empty string for all-JSON input', () => {
-    const input = '{"type": "system"}\n{"type": "result"}';
-    expect(extractCleanOutput(input)).toBe('');
-  });
-
-  it('handles combined ANSI + JSON + CRLF', () => {
-    const input = '{"type": "system"}\r\n\x1b[33mwarning\x1b[0m\r\n{"type": "result"}';
-    expect(extractCleanOutput(input)).toBe('warning');
-  });
-});
-
-// ========== estimateTokensFromOutput ==========
-
-describe('estimateTokensFromOutput', () => {
-  it('estimates tokens as length / 4', () => {
-    expect(estimateTokensFromOutput('12345678')).toBe(2);
-    expect(estimateTokensFromOutput('1234')).toBe(1);
-  });
-
-  it('floors the result', () => {
-    expect(estimateTokensFromOutput('12345')).toBe(1); // 5/4 = 1.25 → 1
-    expect(estimateTokensFromOutput('123')).toBe(0);   // 3/4 = 0.75 → 0
-  });
-
-  it('returns 0 for empty string', () => {
-    expect(estimateTokensFromOutput('')).toBe(0);
-  });
-});
-
-// ========== extractModifiedFiles ==========
-
-describe('extractModifiedFiles', () => {
-  it('extracts files from "wrote" pattern', () => {
-    const output = 'wrote file "src/index.ts" successfully';
-    expect(extractModifiedFiles(output)).toContain('src/index.ts');
-  });
-
-  it('extracts files from "modified" pattern', () => {
-    const output = 'modified utils.js in place';
-    expect(extractModifiedFiles(output)).toContain('utils.js');
-  });
-
-  it('extracts files from "created" pattern', () => {
-    const output = "created file 'new-file.tsx'";
-    expect(extractModifiedFiles(output)).toContain('new-file.tsx');
-  });
-
-  it('extracts files from "edited" pattern', () => {
-    const output = 'edited config.json';
-    expect(extractModifiedFiles(output)).toContain('config.json');
-  });
-
-  it('deduplicates files', () => {
-    const output = 'wrote src/index.ts\nmodified src/index.ts';
-    const files = extractModifiedFiles(output);
-    expect(files.filter(f => f === 'src/index.ts')).toHaveLength(1);
-  });
-
-  it('returns empty array when no files found', () => {
-    expect(extractModifiedFiles('no files mentioned here')).toEqual([]);
-  });
-
-  it('extracts multiple different files', () => {
-    const output = 'wrote src/a.ts\ncreated src/b.ts\nedited src/c.ts';
-    const files = extractModifiedFiles(output);
-    expect(files).toContain('src/a.ts');
-    expect(files).toContain('src/b.ts');
-    expect(files).toContain('src/c.ts');
-  });
-});
-
-// ========== detectErrorInStderr ==========
-
-describe('detectErrorInStderr', () => {
-  it('detects auth errors', () => {
-    const result = detectErrorInStderr('Error: not logged in to Claude');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('AUTH_REQUIRED');
-  });
-
-  it('detects session expired', () => {
-    const result = detectErrorInStderr('Your session has expired, please re-authenticate');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('AUTH_REQUIRED');
-  });
-
-  it('detects account not found', () => {
-    const result = detectErrorInStderr('account not found for this user');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('ACCOUNT_NOT_FOUND');
-  });
-
-  it('detects API key errors', () => {
-    const result = detectErrorInStderr('invalid api key provided');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('API_KEY_INVALID');
-  });
-
-  it('detects quota exceeded', () => {
-    const result = detectErrorInStderr('quota exceeded for your subscription');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('QUOTA_EXCEEDED');
-  });
-
-  it('detects billing issues', () => {
-    const result = detectErrorInStderr('payment required to continue');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('QUOTA_EXCEEDED');
-  });
-
-  it('detects rate limiting', () => {
-    const result = detectErrorInStderr('rate limit exceeded, retry after 30s');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('RATE_LIMITED');
-  });
-
-  it('detects 429 status', () => {
-    const result = detectErrorInStderr('HTTP 429 too many requests');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('RATE_LIMITED');
-  });
-
-  it('detects network errors', () => {
-    const result = detectErrorInStderr('ECONNREFUSED 127.0.0.1:443');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('NETWORK_ERROR');
-  });
-
-  it('detects DNS failures', () => {
-    const result = detectErrorInStderr('ENOTFOUND api.anthropic.com');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('NETWORK_ERROR');
-  });
-
-  it('detects SSL errors', () => {
-    const result = detectErrorInStderr('CERT_HAS_EXPIRED for api.example.com');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('SSL_ERROR');
-  });
-
-  it('detects service unavailable', () => {
-    const result = detectErrorInStderr('service unavailable, try again later');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('SERVICE_UNAVAILABLE');
-  });
-
-  it('detects 503 status', () => {
-    const result = detectErrorInStderr('HTTP 503 from upstream');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('SERVICE_UNAVAILABLE');
-  });
-
-  it('detects internal errors', () => {
-    const result = detectErrorInStderr('internal server error occurred');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('INTERNAL_ERROR');
-  });
-
-  it('detects context too long', () => {
-    const result = detectErrorInStderr('context too long, exceeds 200k tokens');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('CONTEXT_TOO_LONG');
-  });
-
-  it('detects session not found', () => {
-    const result = detectErrorInStderr('session not found, please create a new one');
-    expect(result).not.toBeNull();
-    expect(result!.errorCode).toBe('SESSION_NOT_FOUND');
-  });
-
-  it('returns null for non-matching stderr', () => {
-    expect(detectErrorInStderr('Processing file...')).toBeNull();
-    expect(detectErrorInStderr('Warning: deprecated API usage')).toBeNull();
-    expect(detectErrorInStderr('')).toBeNull();
-  });
-
-  it('returns user-friendly messages', () => {
-    const result = detectErrorInStderr('not logged in');
-    expect(result).not.toBeNull();
-    expect(result!.message).toContain('authentication');
-    // Should not expose raw error
-    expect(result!.message).not.toContain('not logged in');
-  });
-});

package/server/cli/headless/stall-assessor.test.ts

@@ -1,165 +0,0 @@
-import { describe, expect, it } from 'vitest';
-import type { StallContext } from './stall-assessor.js';
-
-// quickHeuristic, parseAssessmentResponse, and parseVerdictResponse are not exported.
-// We test them via assessStall (which calls quickHeuristic first) and by testing
-// the parsing functions indirectly. Since quickHeuristic is the critical logic
-// and assessStall calls it before Haiku, we can test the heuristic paths by
-// providing contexts that match known patterns.
-//
-// To avoid spawning Haiku (which requires `claude` CLI), we only test contexts
-// that trigger the heuristic fast-path (return non-null from quickHeuristic).
-
-import { assessStall } from './stall-assessor.js';
-
-function makeContext(overrides: Partial<StallContext> = {}): StallContext {
-  return {
-    originalPrompt: 'Fix the bug in auth.ts',
-    silenceMs: 120_000,
-    pendingToolCount: 0,
-    totalToolCalls: 5,
-    elapsedTotalMs: 300_000,
-    ...overrides,
-  };
-}
-
-describe('assessStall - quickHeuristic paths', () => {
-  it('extends when tokens are still flowing (tokenSilenceMs < 60s)', async () => {
-    const ctx = makeContext({ tokenSilenceMs: 30_000 });
-    const verdict = await assessStall(ctx, 'claude', false, false);
-    expect(verdict.action).toBe('extend');
-    expect(verdict.extensionMs).toBe(10 * 60_000);
-    expect(verdict.reason).toContain('Tokens still flowing');
-  });
-
-  it('extends when tokenSilenceMs is 0', async () => {
-    const ctx = makeContext({ tokenSilenceMs: 0 });
-    const verdict = await assessStall(ctx, 'claude', false, false);
-    expect(verdict.action).toBe('extend');
-    expect(verdict.reason).toContain('Tokens still flowing');
-  });
-
-  it('does not use token heuristic when tokenSilenceMs >= 60s', async () => {
-    const ctx = makeContext({
-      tokenSilenceMs: 60_000,
-      pendingToolCount: 3, // will trigger parallel tools heuristic
-    });
-    const verdict = await assessStall(ctx, 'claude', false, false);
-    // Should NOT hit the token heuristic, should hit the 3+ parallel tools one
-    expect(verdict.action).toBe('extend');
-    expect(verdict.reason).toContain('parallel tool calls');
-  });
-
-  it('defers to watchdog when active and tools are pending', async () => {
-    const ctx = makeContext({ pendingToolCount: 1, lastToolName: 'Bash' });
-    const verdict = await assessStall(ctx, 'claude', false, true);
-    expect(verdict.action).toBe('extend');
-    expect(verdict.extensionMs).toBe(15 * 60_000);
-    expect(verdict.reason).toContain('Watchdog active');
-  });
-
-  it('defers to watchdog and lists pending tool names', async () => {
-    const ctx = makeContext({
-      pendingToolCount: 2,
-      pendingToolNames: new Set(['WebFetch', 'Bash']),
-    });
-    const verdict = await assessStall(ctx, 'claude', false, true);
-    expect(verdict.action).toBe('extend');
-    expect(verdict.reason).toContain('WebFetch');
-    expect(verdict.reason).toContain('Bash');
-  });
-
-  it('extends for Task subagent via pendingToolNames', async () => {
-    const ctx = makeContext({
-      pendingToolCount: 1,
-      pendingToolNames: new Set(['Task']),
-    });
-    const verdict = await assessStall(ctx, 'claude', false, false);
-    expect(verdict.action).toBe('extend');
-    expect(verdict.reason).toContain('Task subagent');
-  });
-
-  it('extends for Task subagent via lastToolName fallback', async () => {
-    const ctx = makeContext({
-      pendingToolCount: 1,
-      lastToolName: 'Task',
-    });
-    const verdict = await assessStall(ctx, 'claude', false, false);
-    expect(verdict.action).toBe('extend');
-    expect(verdict.reason).toContain('Task subagent');
-  });
-
-  it('scales Task extension with pending count', async () => {
-    const ctx1 = makeContext({
-      pendingToolCount: 1,
-      pendingToolNames: new Set(['Task']),
-    });
-    const ctx3 = makeContext({
-      pendingToolCount: 3,
-      pendingToolNames: new Set(['Task']),
-    });
-    const v1 = await assessStall(ctx1, 'claude', false, false);
-    const v3 = await assessStall(ctx3, 'claude', false, false);
-    // More pending = more extension, capped at 30 min
-    expect(v3.extensionMs).toBeGreaterThanOrEqual(v1.extensionMs);
-    expect(v3.extensionMs).toBeLessThanOrEqual(30 * 60_000);
-  });
-
-  it('extends for 3+ parallel tool calls', async () => {
-    const ctx = makeContext({ pendingToolCount: 3 });
-    const verdict = await assessStall(ctx, 'claude', false, false);
-    expect(verdict.action).toBe('extend');
-    expect(verdict.extensionMs).toBe(15 * 60_000);
-    expect(verdict.reason).toContain('parallel tool calls');
-  });
-
-  it('extends for 5 parallel tool calls', async () => {
-    const ctx = makeContext({ pendingToolCount: 5 });
-    const verdict = await assessStall(ctx, 'claude', false, false);
-    expect(verdict.action).toBe('extend');
-    expect(verdict.reason).toContain('5 parallel tool calls');
-  });
-
-  it('extends for WebSearch without watchdog', async () => {
-    const ctx = makeContext({ lastToolName: 'WebSearch', pendingToolCount: 1 });
-    // pendingToolCount < 3, not Task, not watchdog active, but WebSearch
-    const verdict = await assessStall(ctx, 'claude', false, false);
-    expect(verdict.action).toBe('extend');
-    expect(verdict.extensionMs).toBe(5 * 60_000);
-    expect(verdict.reason).toContain('WebSearch');
-  });
-
-  it('extends for WebFetch without watchdog', async () => {
-    const ctx = makeContext({ lastToolName: 'WebFetch', pendingToolCount: 1 });
-    const verdict = await assessStall(ctx, 'claude', false, false);
-    expect(verdict.action).toBe('extend');
-    expect(verdict.extensionMs).toBe(5 * 60_000);
-    expect(verdict.reason).toContain('WebFetch');
-  });
-
-  it('does NOT extend for WebSearch when watchdog is active', async () => {
-    // When watchdog is active and tools are pending, the watchdog deferral
-    // takes priority over the WebSearch heuristic
-    const ctx = makeContext({
-      lastToolName: 'WebSearch',
-      pendingToolCount: 1,
-    });
-    const verdict = await assessStall(ctx, 'claude', false, true);
-    // Should defer to watchdog, not WebSearch heuristic
-    expect(verdict.action).toBe('extend');
-    expect(verdict.reason).toContain('Watchdog active');
-  });
-
-  it('falls back to extend when Haiku assessment fails', async () => {
-    // Context that doesn't match any heuristic → triggers Haiku →
-    // Haiku fails (no `claude` binary) → cautious extend
-    const ctx = makeContext({
-      pendingToolCount: 1,
-      lastToolName: 'Edit',
-    });
-    const verdict = await assessStall(ctx, 'nonexistent-claude-binary', false, false);
-    expect(verdict.action).toBe('extend');
-    expect(verdict.extensionMs).toBe(10 * 60_000);
-    expect(verdict.reason).toContain('unavailable');
-  });
-});