Repolode
Submit
tophant-ai

tophant-ai/ClawVault

A security vault for OpenClaw and Claude Code that detects secrets, prompt injection and dangerous commands in AI traffic and keeps audit logs.

View on GitHub
Stars
1,072
Forks
37
Open issues
4
Last commit
2 months ago

What is it?

ClawVault is a local security layer for AI agents. It runs a transparent proxy gateway, local agent hooks and a web dashboard to watch traffic sent to AI model providers, OpenClaw sessions, and Claude Code prompts and tool calls.

The detection engine looks for API keys, passwords, PII and credit card numbers, as well as prompt injection patterns such as role hijacking and instruction override. A guard layer can allow, block or sanitize each request, replacing secrets with placeholders. Dangerous shell commands such as rm -rf or curl piped to bash can be intercepted, and daily or monthly token budgets can be enforced.

All events are stored in a local SQLite audit log and shown in the dashboard, which also provides a protection monitor and per-agent configuration. Guard behaviour can be set to strict, interactive or permissive.

Key features

  • 01Detects API keys, passwords, PII and credit card numbers in prompts and traffic.
  • 02Blocks prompt injection such as role hijacking and data exfiltration attempts.
  • 03Intercepts dangerous shell commands like rm -rf and curl piped to bash.
  • 04Replaces detected secrets with placeholders before requests reach the model.
  • 05Provides a Claude Code PreToolUse and UserPromptSubmit hook integration.
  • 06Records audit logs in SQLite with a web dashboard for review.

Use cases

  • Keeping API keys and private files out of prompts sent to AI agents.
  • Auditing what an OpenClaw or Claude Code agent sends and runs.
  • Setting daily token budgets on AI provider usage.
  • Blocking dangerous commands issued by coding agents on a development machine.

Topics

#ai-security#ai-agents#prompt-injection#dlp#audit-log#python#claude-code#llm#clawdbot#clawdbot-plugin

FAQ

What is ClawVault?
A security vault for OpenClaw and Claude Code that detects secrets, prompt injection and dangerous commands in AI traffic and keeps audit logs. ClawVault is a local security layer for AI agents. It runs a transparent proxy gateway, local agent hooks and a web dashboard to watch traffic sent to AI model providers, OpenClaw sessions, and Claude Code prompts and tool calls.
Is ClawVault free to use? What license does it have?
ClawVault is open source under the MIT license. Check the license terms before using it in commercial products.
Is ClawVault still maintained?
Updated within 6 months. The last commit to ClawVault was on Aug 20, 2026.
What can I use ClawVault for?
Keeping API keys and private files out of prompts sent to AI agents. Auditing what an OpenClaw or Claude Code agent sends and runs. Setting daily token budgets on AI provider usage. Blocking dangerous commands issued by coding agents on a development machine.
Where can I get ClawVault?
The source code and installation instructions are on GitHub: https://github.com/tophant-ai/ClawVault

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects