Add detection patterns for new supply chain attack techniques

[azrollin]

After scanning the real axios RAT (North Korean Lazarus group), we added 8 supply chain patterns. But there are more attack techniques in the wild:

• Typosquatting detection (similar package names)
• Dependency confusion patterns
• Build script hijacking (Makefile, setup.py tricks)
• Git hook injection

Pick one technique, write the detection pattern, add a test. See sunglasses/attacks/supply_chain.py for format.

Report: sunglasses.dev/report-axios-rat.html

[azrollin]

Update (v0.2.3, April 4, 2026):

Significant progress since this issue was created. Here's what was added:

v0.1.1 (April 1): 8 supply chain attack patterns including:

• Dependency confusion detection
• Typosquatting indicators
• Build script hijacking (setup.py, Makefile)
• Post-install hook abuse
• Package name similarity checks

v0.2.0–v0.2.2:

• New agent_workflow_security category (6 patterns) covering web-fetch-to-publish pipeline injection, overprivileged connectors, poisoned persistent prompts, RSS/brand doc ingestion poisoning, and unsafe auto-publish detection
• Total patterns now at 75 (up from 53), 442 keywords, 18 categories

Real-world validation: Scanned the actual North Korean Lazarus/BlueNoroff axios RAT — caught 3 threats in 3.67ms. Report: sunglasses.dev/report-axios-rat.html

Still looking for contributions on:

• Git hook injection patterns (.git/hooks/ abuse)
• Lock file manipulation detection (package-lock.json, yarn.lock poisoning)
• CI/CD pipeline injection (GitHub Actions, GitLab CI config tampering)
• Registry confusion attacks (private vs public registry resolution)
• More real-world supply chain attack samples to validate against

The attack-db now has structured JSON entries under attack-db/attacks/ — easy format to contribute to. 🕶️