ai-safety — independent software & tools
-
cavra
CAVRA - Controlled Agentic Verification & Runtime Authority. Before the agent acts, CAVRA decides. CAVRA is a runtime governance and authority layer for AI coding agents.
-
hyperspace-terminal
Hyper Terminal 2026: Next-Gen Web Tech CLI ⚡ | Free & Open Source
-
codex-safe-starter
Safe repo-local starter template for Codex guidance, skills, read-only agents, conservative config, and no-key validation.
-
anywhere-agents
Manage agent configurations across projects with a single AGENTS.md file to standardize rules, skills, and permissions for any LLM-powered tool.
-
sophia-agi
Wisdom before intelligence. A provenance-aware reasoning layer that abstains instead of fabricating — stops LLMs inventing attributions and merging distinct traditions. Measured: 0% fabrication on unknown-answer traps (vs 17–25% raw); −12.5pt hallucinated attributions at 0% false-positive cost. An AGI-candidate proof package, not proven AGI.
-
outlier-audit
AI coding audit, token tracker, and deskilling detector. Local-first CLI that measures AI authorship, carbon footprint, agent blast radius, and context waste — with warnings, not surveillance.
-
AgentGuard
Protect AI agents by detecting and blocking prompt, command injection, Unicode bypass, and social engineering attacks with customizable security controls.
-
Secure-Agent-Launcher
Block AI agent access to sensitive macOS paths and log all actions to protect private data during command execution.
-
hallucinate.md
The open standard for telling AI not to hallucinate.
-
pattern8
Enforce zero-trust rules for AI agents to prevent hallucinations, unsafe actions, and policy bypasses
-
bonklm
Enforce security guardrails for large language models in Node.js applications to manage risks and maintain safe usage across platforms and providers.
-
Claude-cod-top-2026
Still trusting results that AI generated, tested, and declared successful by itself? This Claude Code methodology adds evidence markers, audit gates, anti-hallucination rules, and Validation Theater detection to keep AI-assisted development verifiable.