v0.3.1

Security

• Path traversal fix: brand_audit_content, brand_check_compliance, and brand_audit_drift accepted file paths without cwd validation. An agent could read any .html/.md/.txt file on the filesystem. Now all three resolve paths relative to cwd and reject escapes.
• SSRF bypass fix: brand_extract_messaging used bare fetch() instead of safeFetch(), bypassing all SSRF protection. Now all outbound HTTP goes through safeFetch().
• Response size limits: HTML responses capped at 5MB, CSS at 1MB per stylesheet. Prevents memory exhaustion from malicious URLs.
• Feedback rate limiting: Max 5 entries/minute, 100 total files. Prevents disk exhaustion from agent flooding.