Inner Warden

Open-source runtime guardrail for AI agents: screens shell commands and MCP/tool calls before they run (allow/review/deny). Cross-OS, Apache-2.0.

Point your AI coding agent at this repo and it installs, wires and verifies InnerWarden, the runtime guardrail that screens an agent's commands before they run.

InnerWarden release binaries + install scripts (source is private; source-available to customers under licence)

Open detection rules for AI agent threats — like Sigma, but for LLMs. 71 YAML rules across 9 categories.

Attack one server, all others defend. Collaborative threat sharing for Inner Warden with cryptographic trust and auto-expiring blocks.

Identify attackers by methodology, not IP — behavioral fingerprinting with order-preserving hashes and anomaly detection

8 attack patterns in 9 bits — kernel-level kill chain detection using syscall bitmasks and eBPF

Reproducible isolated lab for verifying Inner Warden's detection of public Linux kernel CVEs. Clone, follow the per-CVE recipe, watch the autonomous agent react on your own VM.

Ring -2 firmware security — SMM monitoring, MSR guards, SPI flash integrity, UEFI attestation

Detect firmware rootkits by measuring time. CPU cycle-accurate timing attestation for x86 and ARM — no hardware required.