YggLucidAkshay/kavach
↗ GitHubTactical AI Workspace Monitor & EDR
234
Stars
42
Forks
1
Watchers
0
Open Issues
Safety Rating A
Kavach appears to be a legitimate open-source security tool. Its capabilities (process termination, clipboard manipulation, file redirection) are powerful but are clearly documented as intentional EDR defense mechanisms. No hardcoded secrets, obfuscated code, data exfiltration patterns, or prompt injection attempts were found in the provided content. The project is GPLv3-licensed and authored transparently. The elevated system privileges requested (Administrator on Windows, Full Disk Access on macOS) are consistent with the stated security monitoring purpose.
ℹAI-assisted review, not a professional security audit.
AI Analysis
Kavach (Sanskrit for 'Armor') is an open-source Endpoint Detection and Response (EDR) desktop application built with Tauri v2, Rust, and React, designed to monitor, restrain, and remediate autonomous AI agents and local LLMs operating on a local machine. It provides a userland file system observer that intercepts destructive operations, routes them to a phantom directory, and offers forensic tools including cryptographic audit logs, temporal rollback, honeypot tripwires, clipboard entropy analysis, PII sanitization, and supply chain CVE scanning against package.json dependencies.
Use Cases
- Monitoring autonomous AI agent activity on a local workstation to detect and quarantine destructive file operations
- Protecting sensitive files and secrets from rogue or hallucinating LLM agents operating locally
- Forensic auditing of AI agent actions via an immutable cryptographic log chain
- Detecting and preventing clipboard-based secret exfiltration during AI agent sessions
- Scanning workspace dependencies for known malicious or vulnerable packages in real time
Tags
Security Findings (4)
No hardcoded secrets, API keys, or tokens were detected in the provided repository content. The README references a honeypot decoy file 'system_auth_tokens.json' which is described as synthetic bait data, not real credentials.
No malicious code patterns detected. Features such as OS process termination (WMIC, renice), clipboard overwriting, and child process restriction are documented as intentional security defense mechanisms and appear consistent with the stated EDR purpose.
No manifest files were provided for static analysis. The README mentions a built-in supply chain auditor that scans workspace package.json for CVEs, but no dependency manifests for Kavach itself were available to assess.
No prompt injection attempts detected in the README or metadata. Content is straightforwardly descriptive of the application's features and architecture.
Project Connections
skyclaw
→Kavach is purpose-built to monitor and contain autonomous AI agents like those run by skyclaw (TEMM1E). Kavach's EDR layer could protect a workstation hosting a skyclaw agent runtime, intercepting any destructive file operations the agent might perform.
clawvault
→ClawVault stores AI agent memory and context as local markdown files. Kavach could monitor the ClawVault workspace directory to detect and quarantine any unauthorized modifications or reads of sensitive memory files by rogue agents.