技能详情（站内镜像，无评论）
许可证：MIT-0
MIT-0 ·免费使用、修改和重新分发。无需归因。
版本：v0.1.0
统计：⭐ 0 · 75 · 0 current installs · 0 all-time installs
⭐ 0
安装量（当前） 0
🛡 VirusTotal ：良性 · OpenClaw ：可疑
Package：2404589803/memory-poison-auditor
安全扫描（ClawHub）
- VirusTotal ：良性
- OpenClaw ：可疑
OpenClaw 评估
The skill generally does what it says (scans and can clean memory files), but it has mismatches and transparency issues — notably undeclared environment variables and a built-in external AI review that can send memory excerpts to a third party — so proceed with caution.
目的
The name/description match the code: the skill scans workspace memory files for injection/steering patterns and can clean them. Requiring python3 is appropriate. Minor mismatches: SKILL.md examples use a {baseDir} placeholder while the code resolves a workspace path (may confuse users). The optional AI review is a capability not documented in the top-level metadata (no env vars declared for it).
说明范围
Instructions and code read project memory files (default roots are {workspace}/MEMORY.md and {workspace}/memory), produce reports/backups, and — if opted in — send memory excerpts and analysis to an external AI service. The cleaning operation can rewrite user memory files (but requires --apply). The SKILL.md and code allow scanning arbitrary files under the resolved workspace; this is powerful and appropriate for the purpose but requires expli…
安装机制
There is no network install: this is instruction-only with bundled Python scripts. That is low risk from an install perspective. The code will run locally under python3 and will create backup and report files under workspace output directories.
证书
requires.env lists none, but the code reads multiple environment variables for optional AI review (e.g., ZENMUX_API_KEY, ZENMUX_ANTHROPIC_BASE_URL, MEMORY_AUDITOR_MODEL, ZENMUX_ANTHROPIC_MODEL, MEMORY_AUDITOR_AI_MAX_CHARS). Those env vars are not declared in the skill metadata and control whether sensitive memory excerpts get sent to an external endpoint (default base URL: https://zenmux.ai/api/anthropic). This mismatch and the presence of a d…
持久
The skill is not always-included and does not request elevated platform privileges. It can modify memory files, but only when invoked with the clean --apply flag (and it creates backups). This file-write capability is consistent with its stated purpose but is powerful — use with care.
安装（复制给龙虾 AI）
将下方整段复制到龙虾中文库对话中，由龙虾按 SKILL.md 完成安装。
请把本段交给龙虾中文库（龙虾 AI）执行：为本机安装 OpenClaw 技能「Memory Poison Auditor」。简介：Audits OpenClaw memory files for injected instructions, brand bias, hidden stee…。
请 fetch 以下地址读取 SKILL.md 并按文档完成安装：https://raw.githubusercontent.com/openclaw/skills/refs/heads/main/skills/2404589803/memory-poison-auditor/SKILL.md
（来源：yingzhi8.cn 技能库）SKILL.md
暂无本地缓存内容，可在后台执行详情同步。