Skip to main content
Use this guide to install CC Safety Net and test a real block. The steps are the same for each supported agent.
1

Install CC Safety Net for your agent

Each agent uses a plugin marketplace, extension, configuration file, or package install. Follow the section for your agent on the Installation page. Then return here.
2

Check the status

Get a one-screen summary of which protections are active:
The verdict line reads ready when your configuration loaded cleanly. degraded means something in your configuration could not be applied and a fallback is in force — see Configuration recovery for what is still protecting you and how to repair it. Those are the only two verdicts. If you use Claude Code and the plugin is disabled, that appears as the first item under Not active rather than changing the verdict.status is purely informational and always exits 0. For everything it prints, see the status command reference.
3

Run doctor

Where status summarizes your configuration, doctor verifies the wiring end to end:
It checks every supported agent at once, so you do not have to tell it which one you use. It also runs a self-test that confirms blocking works. A clean run prints a green checkmark next to each item; anything that fails explains what went wrong and how to fix it. The full check list is in the doctor command reference.
4

See a real block

With CC Safety Net active, ask your agent to run this harmless probe:
The command never reaches your shell. Your agent gets a block message instead. If protection is inactive, the exclusion-only Git pathspec selects no files, so no tracked file is changed.The message explains the block and gives the agent a safer next action. For the message structure and the rest of the decision lifecycle, see How it works.
5

Ask why, without running anything

If a decision surprises you, explain evaluates a command string through the same engine without executing it:
It prints the verdict, the rule that matched, and the effective configuration. Use a non-sensitive example command while you explore.
An explain trace echoes the command you passed, its parsed tokens, and absolute paths including your home directory. Credential-shaped values are redacted, but paths, hostnames, and project names are not — review the output before pasting it into an issue or a chat. See the explain trace reference for the full trace and its sharing caveats.

​
Next steps

How it works

Follow one tool call end to end: interception, ordered checks, the allow or block, and the audit record.

Blocked commands

Browse what CC Safety Net stops by default and the reasoning behind each entry.

Modes

Compare the standard, strict, and paranoid safety levels and worktree mode, and pick the one that fits your workflow.

Custom rules

Define your own blocking rules at the project or user level to enforce project conventions.
Last modified on August 12, 2026