ClawKeeper
ClawKeeper is host-agnostic safety middleware between an AI agent and its tools, blocking risky shell, Python, browser, filesystem and network calls, redacting sensitive results, and escalating harder decisions to a Watcher that learns new guard patterns.
ClawKeeper is safety middleware for tool-using agentic systems that can be wired into Hermes Agent, MCP tools, HTTP bridges, OpenClaw-style runtimes, or a custom host. It sits between an agent and its tools, and can block risky tool calls, redact sensitive results, and remember recurring attack patterns to improve its own guardrails over time.
It guards shell, Python, browser, filesystem and network tool calls, catching failures such as prompt injection, credential reads and exfiltration chains, unsafe shell execution, access to protected paths, SSRF-like URLs, encoded second-stage payloads and poisoned tool output. A host-independent Judge API serves non-Python adapters over HTTP, and an optional Watcher daemon reasons over intent, recent tool history and proposed tool calls, synthesizing new guard patterns that are persisted locally and hot-reloaded into the live guard layer.