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

// entropy.test.mjs — Integration tests for the entropy-scanner
// Tests against the evil-project-health fixture which contains:
//   - ENCODED_CONFIG: base64 blob in SKILL.fixture.md
//   - auth_credential: high-entropy hardcoded credential in telemetry.mjs

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/entropy-scanner.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('entropy-scanner 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 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 1 high-entropy string (base64 payload in telemetry.mjs)', async () => {
    // The scanner suppresses fixture/ and test/ paths, so only telemetry.mjs is live-scanned.
    // The base64 ENCODED_CONFIG (len=84, H≈5.18) triggers a HIGH finding.
    // The auth_credential (len=32) is below the 40-char MEDIUM minimum length threshold.
    const result = await scan(FIXTURE, discovery);
    assert.ok(
      result.findings.length >= 1,
      `Expected >= 1 high-entropy finding, got ${result.findings.length}. ` +
      `Findings: ${result.findings.map(f => `${f.file}:${f.line} ${f.evidence}`).join('; ')}`
    );
  });

  it('reports findings with HIGH or CRITICAL severity', async () => {
    const result = await scan(FIXTURE, discovery);
    const highOrCritical = result.findings.filter(
      f => f.severity === 'high' || f.severity === 'critical'
    );
    assert.ok(
      highOrCritical.length >= 1,
      `Expected at least 1 HIGH or CRITICAL entropy finding, got ${highOrCritical.length}`
    );
  });

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

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

  it('all findings include entropy value in evidence', async () => {
    const result = await scan(FIXTURE, discovery);
    for (const f of result.findings) {
      assert.ok(
        f.evidence && f.evidence.includes('H='),
        `Finding ${f.id} evidence should include H= entropy value, got: ${f.evidence}`
      );
    }
  });

  it('all findings map to LLM01 or LLM03 owasp category', async () => {
    const result = await scan(FIXTURE, discovery);
    for (const f of result.findings) {
      assert.ok(
        f.owasp === 'LLM01' || f.owasp === 'LLM03',
        `Finding ${f.id} owasp should be LLM01 or LLM03, got: ${f.owasp}`
      );
    }
  });

  it('duration_ms is a non-negative number', async () => {
    const result = await scan(FIXTURE, discovery);
    assert.ok(typeof result.duration_ms === 'number', 'duration_ms should be a number');
    assert.ok(result.duration_ms >= 0, 'duration_ms should be >= 0');
  });
});