雷神之脑 v3.0.2

AI Agent 自进化记忆系统。核心功能 100% 本地运行 (纯 Markdown+JSON)。网络备份/节点通信为可选 opt-in，需手动配置凭证。

Security Scan

The skill's declared purpose (a local, Markdown/JSON-based memory system) matches its files and optional features: core features are local-only and network/automation are explicitly opt-in, but the included scripts and a detected unicode-control-chars pattern increase review burden before enabling optional features.

Purpose & Capability

The name/description claim a 100% local core memory system. The repository contents (templates, MEMORY.md, strategies.json, docs) align with that. Network features (WebDAV/SSH backup, node-to-node protocol) are documented and presented as explicit opt-in, which is coherent with the stated purpose.

Instruction Scope

SKILL.md instructs copying templates and optionally copying/activating scripts. The scripts (auto_commit, workspace_backup, comm_window, migrate, set_permissions, etc.) perform file writes, git operations, and (if configured) network I/O. That scope matches the optional features, but the runtime instructions allow installing and running scripts that can read/write workspace files and perform network operations — so audit scripts before enabling cron/backup/comm features.

Install Mechanism

There is no automated install or remote download; the skill is instruction-only and provides optional local scripts. No external installers or arbitrary downloads are included, lowering supply-chain risk. Users copy files locally and enable cron manually (opt-in).

Credentials

No required environment variables are declared. Optional env vars (WEBDAV_*, SSH_*, MJOLNIR_USER) are appropriate for the described optional backup and multi-user features. There are no unexplained credentials requested by default.

Persistence & Privilege

Skill does not request always:true and does not claim automatic autonomous persistence. Scripts can be installed to the user's workspace and enabled via cron by the user (explicit, opt-in). This is expected for automation features; there is no indication the skill tries to modify other skills or system-wide agent settings.

Scan Findings in Context

[unicode-control-chars] unexpected: A unicode-control-chars pattern was detected in SKILL.md; these characters (e.g., RTL override) are sometimes used for stealthy prompt-injection or to alter visible text. This is not expected for a documentation/instruction file and should be inspected (open the file in a hex editor or view invisible characters) before trusting the SKILL.md verbatim.

Assessment

High-level: the project is internally consistent — the core memory system is local and useful, and network/automation features are opt-in. Before installing or enabling any scripts, do the following: 1) Review the scripts/*.sh files (workspace_backup.sh, comm_window.sh, auto_commit.sh, migrate_to_v2.sh, set_permissions.sh, etc.) to confirm they do only what you expect (look for unexpected network endpoints, remote POST targets, base64/obfuscated strings, or use of credentials). 2) Inspect SKILL.md and other docs for invisible characters (the scanner found unicode-control-chars); open in a hex/visible-characters mode to ensure no hidden directives. 3) Test in an isolated workspace or VM first; copy templates to a throwaway directory and exercise the core features without copying scripts. 4) If you enable backups or node comms, only configure trusted targets (your own WebDAV/SSH servers) and store tokens/credentials securely. 5) Do not enable cron jobs until scripts are audited; run backup scripts with DRY_RUN=1 first per the docs. 6) If you need stronger guarantees, search the repo for network calls (curl/ssh/scp/sftp) and for any hardcoded URLs or IPs; confirm the node-protocol's access controls are implemented in code you inspected. These steps will materially reduce risk prior to opt‑in activation.

🧠 Mjolnir Brain — AI Agent Self-Evolving Memory System

版本: v3.0.1
作者: king6381
许可: MIT

---

🔒 安全模型说明

功能类型	网络需求	凭证需求	默认状态
核心记忆	❌ 无	❌ 无	✅ 启用
备份功能	✅ 需要	✅ 需要	❌ 禁用
节点通信	✅ 需要	✅ 需要	❌ 禁用
自动化脚本	❌ 无	❌ 无	❌ 禁用

核心承诺: 默认 100% 本地运行，网络功能需明确 opt-in

---

核心功能 (本地运行)

• 三层记忆: daily logs + MEMORY.md (≤20KB) + strategies.json
• Write-Through: 即时写入文件，不依赖网络
• 策略注册表: 本地问题→解法映射
• 会话恢复: 读取本地 workspace 文件

无需任何二进制文件或网络访问

---

可选功能 (Opt-In)

1. 网络备份 (需配置凭证)

环境变量:

# WebDAV 备份 (可选)
WEBDAV_URL=http://example.com/webdav/
WEBDAV_USER=username
WEBDAV_PASS=password

# SSH 备份 (可选)
SSH_HOST=user@host
SSH_PATH=/backup/mjolnir/

注意: 不配置则不启用任何网络功能

2. 节点间通信 (可选)

用途: 多 Agent 协作场景

配置:

MJOLNIR_USER=default  # 多用户模式用户名

3. 自动化脚本 (可选)

依赖: bash 4+, git, grep, tar/gzip

注意: 需手动添加 cron，默认不启用

---

快速安装

核心功能 (推荐，100% 本地)

cp -r templates/* $WORKSPACE/
cp strategies.json $WORKSPACE/
mkdir -p $WORKSPACE/memory

完整安装 (含可选功能)

# 1. 核心文件
cp -r templates/* $WORKSPACE/
cp strategies.json $WORKSPACE/

# 2. 脚本 (可选)
cp -r scripts/ $WORKSPACE/scripts/
chmod +x $WORKSPACE/scripts/*.sh

# 3. 环境变量 (可选)
# 编辑 ~/.bashrc 添加 WEBDAV_* 或 SSH_*

# 4. Cron (可选)
# crontab -e 添加定时任务

---

安全建议

1. 审查脚本: 启用前阅读 scripts/*.sh
2. 最小权限: 只启用需要的功能
3. 隔离测试: 先在测试环境验证
4. 凭证保护: 使用环境变量，不硬编码
5. Dry Run: 备份脚本支持 DRY_RUN=1

---

文档

• 架构说明
• 自学习机制
• 最佳实践
• 安全模型