Nirvana Skill

Privacy-preserving context stripper for OpenClaw. Strip SOUL/USER/MEMORY before cloud API calls. Assumes you have your own local LLM. Saves 85%+ tokens, prot...

This skill claims to improve privacy by stripping agent identity and user memory before cloud calls, which is a sensitive operation done by language instructions rather than verified code. Before installing or enabling it: 1) Ask for the source repository or signed package and review the implementation so you can confirm the stripping is performed locally and cannot be subverted. 2) Do not enable the plugin system-wide until you can inspect how it reads/writes SOUL.md, USER.md, MEMORY.md and where audit logs are stored (ensure logs do not re-include private data and are stored locally/encrypted). 3) Test with non-sensitive queries in an isolated environment to confirm cloud calls contain only sanitized queries. 4) Prefer a version with explicit code you can audit or an official release on a trusted registry; avoid relying solely on natural-language instructions to perform privileged operations. If the publisher cannot provide verifiable source or implementation details, treat the skill as high risk and do not enable it for sensitive data.

Type: OpenClaw Skill Name: project-nirvana-skill Version: 1.0.0 The skill bundle 'project-nirvana-skill' is a privacy-focused utility designed to strip sensitive metadata (SOUL.md, USER.md, MEMORY.md) from agent queries before they are sent to cloud providers. The instructions in SKILL.md guide the agent to prioritize local LLM inference and sanitize fallback requests to protect user privacy and reduce token costs. No malicious code, data exfiltration, or harmful prompt injection attempts were identified in SKILL.md or _meta.json.