Overview

DefenseClaw is the Cisco guardrail for AI coding agents. It scans what an agent installs, checks each prompt and tool call, and keeps an audit trail.

DefenseClaw checks what your AI coding agent installs and does. It scans skills and MCP servers, judges each prompt and tool call, blocks the unsafe ones when you ask it to, and records every decision.

Guided example · Pre-authored outcomes

One Claude Code action under three operating modes

Change the mode to see the same HIGH-risk action log, block, or pause through a predefined outcome.

Deterministic
Outcome
mode: actionblock_at: HIGHhuman_approval: truehitl_min_severity: highcritical_behavior: always_blockconnector:  id: claudecode  native_ask_event: PreToolUsemode: actionblock_at: HIGHhuman_approval: truehitl_min_severity: highcritical_behavior: always_blockconnector:  id: claudecode  native_ask_event: PreToolUse
DecisionPause for approval
Reason

Action mode + HITL + native ask support

Action

Wait for operator

02

PauseHITL is enabled and Claude Code supports native ask.

Step 2 / 2
What DefenseClaw did — and did not do

What it did

  • Show deterministic outcomes for one action under three modes
  • Use Claude Code connector capabilities in the result

What it did not do

  • Run a policy engine in the browser
  • Allow observe mode to block
  • Pause a CRITICAL finding

The demo sends one synthetic Claude Code PreToolUse event with a HIGH finding through three modes:

  • Observe lets the call run and records it.
  • Action blocks it. The demo sets the block level to HIGH; the shipped default blocks CRITICAL and alerts on MEDIUM and above.
  • Action with human approval (HITL) pauses the call and asks you inside Claude Code.