PROMPT INJECTION PROTECTION

Protects against prompt injection attacks by sanitizing, validating, and securely processing untrusted external content from websites, emails, and documents.

This package appears to be a coherent prompt-injection protection library. Before installing or enabling it broadly, consider: (1) it persists learned patterns to learned-threats.json in the skill directory — confirm that persistent writes are acceptable in your environment; (2) its constructor starts an auto-update interval (periodic background task) — if you want to avoid background timers, disable auto-updates or adjust the update schedule; (3) although demos reference sensitive files/commands, the code only matches patterns and does not read those files or require credentials; (4) review the source files yourself if you need stronger assurance (especially adaptive-learning.js and auto-update.js) and consider disabling learning in high-security contexts. If you plan to allow autonomous invocation, remember the skill will adapt to inputs it sees, so ensure it runs in a sandboxed agent context and that learned data is acceptable to retain.

Type: OpenClaw Skill Name: prompt-injection-protection Version: 1.0.0 This OpenClaw skill is designed for prompt injection protection, featuring content sanitization, threat detection, adaptive learning, and auto-updates. All code and documentation consistently describe defensive mechanisms against various attack vectors, including command execution, system file access, and prompt manipulation. The demo files explicitly show malicious inputs being *detected and alerted upon* by the protection system, rather than being executed. There is no evidence of intentional harmful behavior, data exfiltration, unauthorized execution, or prompt injection against the agent itself; instead, the skill aims to prevent these actions.