Bouncer — Active Security Layer for Claude Code

Bouncer is the active security layer for Claude Code — watching every prompt, file write, and bash command in real time. It blocks prompt injection, hardcoded secrets, and destructive operations before they happen.

5 active barriers (always on after install):

| Barrier | What it catches | |---------|----------------| | Every prompt | Injection attempts, role overrides, jailbreaks, hidden instructions in pasted content | | Pre-file write | Hardcoded secrets, API keys, passwords, connection strings | | Post-file write | Patterns that slipped through — suspicious blobs, phone-home URLs, backdoor code | | Bash commands | Destructive operations, unauthorized network calls, pipe-to-shell attacks, force-pushes to main | | Code audit | Full security review on demand with /bouncer:scan, /bouncer:secrets, /bouncer:audit |

5 manual commands: /bouncer:scan, /bouncer:secrets, /bouncer:audit, /bouncer:check, /bouncer:injection

2 specialist agents: bouncer:injection-detector, bouncer:code-auditor

Install:

bash setup.sh

Active protection starts immediately — no configuration required.

Includes all 5 individual skills (vs $25 separately — save 20%)

Version: 1.0.0 | Ecosystem: OpenClaw / ClawMart