Stories
Step-by-step walkthroughs for common goals — block a destructive command on Claude Code, catch a prompt injection on Codex, stop a secret upload from Cursor, approve risky tool calls, run the local observability stack, and add a connector without losing audit history.
Six walkthroughs. Each one starts from a goal and ends with an outcome you can check.
Stop Claude Code from deleting a critical path
Watch Claude Code in observe mode, then block rm -rf against a protected path with the bundled CMD-RM-RF rule.
Catch a prompt injection on Codex
Block a prompt injection in Codex with the bundled TRUST-IGNORE-PREVIOUS rule, then add the LLM judge for subtler cases.
Stop a secret upload from Cursor
Block a command that pipes a key file into curl, using Cursor's beforeShellExecution hook and a fake key.
Approve risky tool calls before they run
Turn on human approval (HITL) so a confirmable HIGH-severity tool call pauses in the agent and waits for your yes or no.
Run the local observability stack
Start the bundled Grafana, Prometheus, Loki and Tempo stack, send DefenseClaw telemetry to it, and tear it down.
Add a connector without losing audit history
Guard a second agent next to the first, replace one with another, or remove one. Audit history is kept.
Related how-to pages: Turn on blocking, Allow a blocked call and Turn off the guardrail.
Command generator
Build a Bash or PowerShell defenseclaw setup guardrail command for any connector. Pick mode, scanner backend, detection strategy, rule pack, HITL behaviour, and every advanced knob; copy the result straight into your terminal or CI pipeline.
Stop Claude Code from deleting a critical path
Wire DefenseClaw into Claude Code, collect representative observe-mode traffic, then safely verify CMD-RM-RF against a path that can't be deleted.