CHANGELOG.md

Changelog

All notable changes to this skill are documented here.

[5.2.0] - 2026-02-09

Added

• Semgrep static analysis as 5th scanning tool
  • Detects injection vulnerabilities, eval/exec misuse, hardcoded secrets in code context
  • Uses p/security-audit and p/owasp-top-ten rulesets
  • --no-git-ignore flag ensures gitignored files are scanned (malicious code can hide there)
  • Fills gap between Trivy (dependency CVEs) and manual grep (simple patterns)
• Semgrep install, run commands, and output parsing guidance
• Semgrep section in report template and all examples
• Semgrep severity mapping (ERROR → HIGH-CRITICAL, WARNING → MEDIUM, INFO → LOW)
• workflow_run checkout vulnerability pattern in heuristics.md (CWE-913, from Semgrep code-security rules)
• Semgrep fallback patterns for when tool is unavailable

Changed

• Workflow step 5 now includes Semgrep alongside Trivy and Gitleaks
• Tool Versions table includes Semgrep 1.x
• All example and template version references updated to v5.2.0
• Version bump to 5.2.0

[5.1.0] - 2026-02-07

Added

• Explicit step 14 in workflow: output full GitHub report URL as final screening output
• CRITICAL instruction: report URL must be bare (no backticks, quotes, or markdown) and on its own line for Cmd+click support in terminals like Ghostty

Changed

• Workflow now has 15 steps (was 14)
• "Saving Reports" section expanded with formatting requirements for clickable URLs
• Version bump to 5.1.0

[5.0.0] - 2026-01-31

Added

• screen.sh — one-command screening launcher script
  • ./screen.sh <url> handles Codespace creation, tool installation, and screening
  • --destroy flag: delete Codespace after screening (full wipe)
  • --fresh flag: force new Codespace (ignore existing)
  • --help with usage and security tradeoff explanation
  • Idempotent provisioning via ~/.screening-tools-installed marker
  • Auto-run: reuses Codespace and starts Claude screening automatically on subsequent runs
  • Fallback: opens interactive SSH if Claude not logged in yet
  • Codespace lookup via --display-name screener
• README: "One-Command Screening" section at top
• README: "Reuse vs Destroy: Security Tradeoffs" section

Changed

• README: file structure updated to include screen.sh
• Version bump to 5.0.0

Deprecated

• screening-launcher skill — replaced entirely by screen.sh

[4.3.2] - 2026-01-29

Fixed

• Report saving uses mv instead of cp — no more duplicate SCREENING-REPORT.md left in repo root
• gh auth login missing -s repo scope — private repo screening failed because default scopes don't include private repo access

[4.3.0] - 2026-01-29

Added

• Screening Private Repos section - gh auth login workflow for private repo access
• Saving Reports section - commit reports to Codespace's repo for GitHub browsing
• Workflow steps 2 (private repo auth) and 13 (save report) added
• Docker/OrbStack gh installation instructions for private repo screening
• README: private repo instructions, report saving instructions
• README: describes skill as Agent Skill with link to agentskills.io

Changed

• Workflow now has 14 steps (was 12)
• README rewritten with installation & usage focus
• Version bump to 4.3.0

[4.2.0] - 2026-01-29

Added

• Deep Dependency Investigation section
  • Install suspicious dependencies in isolation
  • Observe behavior during installation
  • Inspect installed package source code
  • Compare published package vs source repo (supply chain injection detection)
  • Python package investigation workflow
  • Documentation template for investigated dependencies
• New workflow step 10: Deep dive suspicious dependencies
• Deep Dependency Investigation section in report output format

Changed

• Workflow now has 12 steps (was 11)
• Version bump to 4.2.0

[4.1.0] - 2026-01-29

Added

• author field in metadata (gradigit)
• tags for discoverability (7 tags)
• triggers for activation patterns (7 triggers)
• Risk score calculation methodology section
• Tool versions table with tested versions
• Example 6: Tool Installation Failure Recovery
• Copyright holder to LICENSE file

Changed

• Workflow section now explicitly references TaskCreate
• Self-evolution section enhanced with 5 evolution triggers
• Version references updated throughout examples

Audit Result

• Grade: A
• All warnings resolved
• All suggestions implemented

[4.0.0] - 2026-01-29

Added

• Dynamic analysis workflow (execute and observe)
• Security scanning tools: Trivy, Gitleaks, actionlint, zizmor
• Process monitoring during npm/pip install
• Network activity observation
• File system change detection
• Comprehensive risk model section

Changed

• Philosophy: "The sandbox is the protection, not network isolation"
• Network stays connected throughout screening
• Fresh sandbox every time, destroy after use

Removed

• Network disconnect steps (unnecessary with proper sandboxing)
• Paranoid execution restrictions

[3.1.0] - 2026-01-28

Changed

• Replaced Claude.ai web with Docker/OrbStack local sandbox option
• Claude.ai web caused GitHub API rate limiting issues

[3.0.0] - 2026-01-28

Added

• GitHub Codespaces support (cloud sandbox)
• Environment detection commands
• Comparison table: Codespaces vs Docker

[2.1.0] - 2026-01-27

Added

• Environment detection for Codespaces/Docker

[2.0.0] - 2026-01-27

Changed

• Reframed from "audit" to "screening" (accurate naming)
• Added task list workflow for systematic screening

[1.1.0] - 2026-01-26

Added

• Slopsquatting detection (AI-hallucinated packages)
• CVE-2025-30066 (tj-actions/changed-files compromise)
• Expanded token patterns (OpenAI, Anthropic, GitLab, npm, PyPI)

[1.0.0] - 2026-01-25

Added

• Initial release
• Static analysis patterns for malicious code
• Supply chain detection (typosquatting)
• GitHub Actions security checks
• Secret detection patterns
• Prompt injection defense