Security Check

When reviewing code for security, verify the following areas systematically.

1. Injection Patterns

Check that external content is scanned before processing:

• User prompts scanned via UserPromptSubmit hook (blocks at score >= 4)
• Memory writes scanned via createMemCell() injection check
• RAG content scanned before use
• Aho-Corasick scanner covers 44 markers across 7 categories
• Unicode homoglyphs normalized (Cyrillic to Latin)
• Zero-width characters stripped before scanning

2. Sensitive Paths

Verify sensitive path blocking covers:

• .env, .env.* files
• .ssh/* (keys, config, known_hosts)
• Credential files (credentials.json, service-account.json, etc.)
• Token/secret files in common locations
• Check isSensitivePath() and isConfidentialContent() in src/security/blocklist.ts

3. Shell Risk Classification

Verify shell commands are classified correctly:

• CRITICAL: rm -rf /, mkfs, dd if=, format commands
• HIGH: python -c, node -e, docker run, PowerShell -enc, eval(), exec()
• MEDIUM: Package installs, git push, service restarts
• LOW: Read-only operations (ls, cat, git status)
• Check classifyShellRisk() in src/security/policy.ts

4. Output Guard Patterns

Verify LLM output scanning covers:

• XSS: <script>, javascript: URIs, event handlers (onload=, onerror=)
• Code execution: eval(), Function(), import(), child_process
• SQL injection: DROP TABLE, DELETE FROM, UNION SELECT
• Destructive commands: rm -rf /
• Check src/security/output-guard.ts

5. Memory Integrity

Verify memory system is tamper-resistant:

• MemCells have SHA-256 checksums via computeChecksum()
• verifyChecksum() detects tampering
• loadMemCells() validates JSON with isMemCell() type guard
• Injection scanning on write (blocks score >= 4)

6. Authentication and Crypto

Verify cryptographic operations:

• HMAC comparison uses timingSafeEqual with hex validation
• Vault uses XChaCha20-Poly1305 (via @noble/ciphers)
• Vault key file restricted to mode 0o600
• Atomic write-then-rename for vault and cache operations