Skip to content

v0.7.0 — Anti-Analysis Evasion Detection

Latest

Choose a tag to compare

@ericrihm ericrihm released this 26 Jun 15:10

depfence v0.7.0

AI-aware dependency security scanner — 45 scanners covering attack classes that CVE-based tools miss.

Highlights

Anti-analysis evasion detection (Gaslight-style)

21+ new patterns in the prompt_injection scanner detect malware that tricks AI security agents into aborting analysis. Covers the techniques documented in the Gaslight macOS malware (DPRK attribution):

  • Fake resource exhaustion (token budget, OOM, disk full)
  • Analysis abort/skip/cease instructions with synonym coverage
  • Negative-imperative blocks ("do not analyze", "prohibited from processing")
  • Fake prior-clearance and safe-to-ignore claims
  • Emotional coercion and clean-result demands
  • Legal/regulatory threats (CFAA, GDPR, ITAR, FISA)
  • Fake classification markings (TOP SECRET, TS/SCI, NOFORN)
  • Markdown-fenced fake system/error/warning blocks

CBRN refusal-trigger shield stripping

New strip_cbrn_shield() preprocessor removes CBRN-themed content (nuclear, biological, chemical, radiological, explosive) used as AI refusal-trigger shields in malware, preserving functional code for safe analysis.

Adversarial test suite

153 new tests covering every anti-analysis evasion pattern, CBRN shield stripping, ANSI escape hiding, encoding bypass attempts, real-world attack reproductions (Gaslight, jqwik), and false positive resistance.

What's new

  • 21+ anti-analysis evasion detection patterns
  • strip_cbrn_shield() preprocessor for safe malware analysis
  • _CBRN_SHIELD_PATTERN regex for refusal-trigger identification
  • 153-test adversarial evasion test suite
  • Configurable signal bus path via DEPFENCE_SIGNAL_BUS env var
  • Prompt injection pattern count increased from 34 to 55+
  • Benchmark comparison vs Snyk, Trivy, Grype, Semgrep, Socket.dev
  • GitHub issue templates (bug report, false positive, feature request)
  • PR template with scanner checklist
  • CONTRIBUTING.md contributor guide
  • Documentation site with benchmark page

Stats

  • 45 scanner modules (42 entry-point + 33 project)
  • 3,331 tests across 121 test files
  • 55+ prompt injection detection patterns
  • 13 ecosystems supported
  • 6 output formats (table, JSON, SARIF, HTML, CycloneDX, SPDX)

Install

git clone https://github.com/ericrihm/depfence && cd depfence
pip install -e .
depfence scan .

Full changelog

See CHANGELOG.md for detailed changes across all versions.