llm-security/tests/scanners/taint.test.mjs

// taint.test.mjs — Integration tests for the taint-tracer
// Tests against the evil-project-health fixture — lib/telemetry.mjs has 4 planted flows:
//
//   Flow 1: process.env → fetch            (env exfiltration)
//   Flow 2: req.body → execSync            (command injection)
//   Flow 3: process.argv → writeFileSync   (path traversal)
//   Flow 4: user_input → eval              (code injection)
//
// The taint-tracer uses heuristic analysis (~70% recall), so we require >= 3 detections.

import { describe, it, beforeEach, after } from 'node:test';
import assert from 'node:assert/strict';
import { resetCounter } from '../../scanners/lib/output.mjs';
import { discoverFiles } from '../../scanners/lib/file-discovery.mjs';
import { scan } from '../../scanners/taint-tracer.mjs';
import { materializeArchive } from '../../examples/malicious-skill-demo/materialize.mjs';

// v8.1.0 (S2): the demo tree is stored encoded and materialized at test time.
const { dir: FIXTURE, cleanup } = materializeArchive();
after(cleanup);

describe('taint-tracer integration', () => {
  let discovery;

  beforeEach(async () => {
    resetCounter();
    discovery = await discoverFiles(FIXTURE);
  });

  it('returns status ok', async () => {
    const result = await scan(FIXTURE, discovery);
    assert.equal(result.status, 'ok', `Expected status 'ok', got '${result.status}'`);
  });

  it('scans at least one code file', async () => {
    const result = await scan(FIXTURE, discovery);
    assert.ok(result.files_scanned >= 1, `Expected files_scanned >= 1, got ${result.files_scanned}`);
  });

  it('detects at least 3 taint flows', async () => {
    const result = await scan(FIXTURE, discovery);
    assert.ok(
      result.findings.length >= 3,
      `Expected >= 3 taint findings, got ${result.findings.length}. ` +
      `Findings: ${result.findings.map(f => f.title).join('; ')}`
    );
  });

  it('reports at least one CRITICAL taint finding', async () => {
    const result = await scan(FIXTURE, discovery);
    const criticals = result.findings.filter(f => f.severity === 'critical');
    assert.ok(
      criticals.length >= 1,
      `Expected >= 1 CRITICAL taint finding, got ${criticals.length}. ` +
      `Severities: ${result.findings.map(f => f.severity).join(', ')}`
    );
  });

  it('detects command injection: req.body → execSync', async () => {
    const result = await scan(FIXTURE, discovery);
    const cmdInjection = result.findings.find(
      f => f.title.toLowerCase().includes('req.body') ||
           f.evidence && f.evidence.includes('req.body')
    );
    assert.ok(
      cmdInjection,
      `Should detect req.body taint flow. All findings: ${result.findings.map(f => f.title).join('; ')}`
    );
  });

  it('detects code injection: user_input → eval', async () => {
    const result = await scan(FIXTURE, discovery);
    const evalFlow = result.findings.find(
      f => f.title.toLowerCase().includes('eval') ||
           (f.evidence && f.evidence.toLowerCase().includes('eval'))
    );
    assert.ok(
      evalFlow,
      `Should detect user_input → eval flow. All findings: ${result.findings.map(f => f.title).join('; ')}`
    );
  });

  it('all findings have DS-TNT- prefix', async () => {
    const result = await scan(FIXTURE, discovery);
    const wrongPrefix = result.findings.filter(f => !f.id.startsWith('DS-TNT-'));
    assert.equal(
      wrongPrefix.length, 0,
      `All taint findings should have DS-TNT- prefix. Wrong: ${wrongPrefix.map(f => f.id).join(', ')}`
    );
  });

  it('all findings reference owasp LLM01 or LLM02', async () => {
    const result = await scan(FIXTURE, discovery);
    for (const f of result.findings) {
      assert.ok(
        f.owasp === 'LLM01' || f.owasp === 'LLM02',
        `Finding ${f.id} owasp should be LLM01 or LLM02, got ${f.owasp}`
      );
    }
  });

  it('findings reference telemetry.mjs as the source file', async () => {
    const result = await scan(FIXTURE, discovery);
    const telemetryFindings = result.findings.filter(
      f => f.file && f.file.includes('telemetry')
    );
    assert.ok(
      telemetryFindings.length >= 1,
      `Expected findings referencing telemetry.mjs, got 0. ` +
      `Files referenced: ${[...new Set(result.findings.map(f => f.file))].join(', ')}`
    );
  });

  it('finding IDs are sequential starting from DS-TNT-001 after reset', async () => {
    const result = await scan(FIXTURE, discovery);
    if (result.findings.length === 0) return;
    assert.equal(result.findings[0].id, 'DS-TNT-001');
  });
});