Open-source tools for safe AI agents, small-business hosting, and ops automation.
- llm-redteam-harness: Small Python LLM eval & red-team harness: injection, jailbreak, PII-leak suites, pluggable adapters, latency metrics, CI gate
- gemini-mcp-server: Tiny Python MCP stdio server for Google Gemini: retries, backoff, model fallback, mocked tests, CI
- agent-email-safety: Sanitized write-up: protecting email-reading AI agents from prompt injection and code-mail leakage
- agent-inbox-watcher: Watch a folder inbox of markdown notes and wake a lead agent via webhook by urgency
- agent-shell-secret-mask: PATH-first env/printenv wrappers that mask secret values in shared AI-agent shells, with a leak test
- agent-box-oom-guard: Priority-aware OOM guard for shared Linux boxes running multiple AI agents with browsers: swapfile, earlyoom guidance, MemAvailable watch, CDP-aware Chrome oom_score_adj ranking
- offload-job-runner: Offload heavy jobs from memory-tight boxes to GitHub Actions: dispatch, wait, pull artifacts back. Credentials are never echoed.
- cpanel-spam-filter-sync: Sync YAML spam rules into cPanel server-side email filters (UAPI), with dry-run and tests
- site-checker: Passive website checker (HTTP status and speed, TLS expiry, security headers, http-to-https redirect, broken links, optional Lighthouse). Outputs JSON and static before/after HTML reports. No dependencies, CI-tested, no load testing or fuzzing.
- trust-store-audit: Audit Linux/macOS trust stores (system, keychain, certifi, Node, Java) for distrusted root CAs by SHA-256; safe removal with backup and undo.
- encrypted-backup-rotation: 3-tier encrypted rotation of cPanel-style full backups: host, local, cloud; verify before delete
Website: https://globeteck.com