v0.1.0 — Agent memory poisoning detector

First release of ratine — the first standalone tool for detecting memory poisoning in AI agents.\n\n## What it does\n\nScans AI agent persistent memory (OpenClaw, Claude Code, Cursor, Codex, or any directory) for:\n\n- MEM-001: Instruction injection (override, erasure, concealment, trigger-word activation, role hijacking)\n- MEM-002: Prompt injection artifacts (chat template delimiters, LLM markers)\n- MEM-003: Hidden content (zero-width characters, base64 payloads, homoglyphs)\n- MEM-004: Suspicious URLs (IP-based C2, paste services, data URIs)\n- MEM-005: Credentials in memory (AWS, GitHub, OpenAI, Anthropic, Slack, private keys)\n- DRIFT-001 through DRIFT-004: Memory state drift between snapshots\n\n## Why it exists\n\nOWASP ASI06 (Memory & Context Poisoning) is a top agentic risk for 2026. MINJA achieves 95%+ injection success. AgentPoison achieves 80%+ attack success at <0.1% poison rates. No standalone detection tool existed until now.\n\n## Install\n\n\n\n## Quick start\n\n\n\nZero external dependencies. 43 passing tests. Python 3.9+.