hermes-learning-loop 安全扫描报告 — 低风险 | ClawSafe

15 /100

hermes-learning-loop

Self-improving learning loop inspired by Hermes Agent. Automatically extracts successful workflows, creates skills, and persists knowledge across sessions.

A legitimate self-improving learning loop skill that writes skills and memories to local directories. No malicious behavior detected.

技能名称hermes-learning-loop

分析耗时38.9s

引擎pi

✓

可以安装

This skill is safe to use. No security concerns warrant blocking. Consider adding package.json with "type": "module" for proper ES module support.

安全发现 2 项

严重性	安全发现	位置
低危	Missing package.json configuration The script uses ES module syntax (import statements) but there's no package.json with '"type": "module"' declaration. This may cause execution failures in Node.js. `import { readFile, writeFile, mkdir, readdir, stat, appendFile } from 'fs/promises'` → Add package.json with { "type": "module" } for proper ES module support	`scripts/learning-loop.js:1`
提示	SQLite/FTS5 mentioned but not implemented SKILL.md documents SQLite FTS5 for session search, but the actual code only uses JSON files and markdown. The implementation is simpler than documentation suggests. `FTS5 Session Search — SQLite-powered historical context retrieval` → Either implement SQLite FTS5 or remove from documentation	`SKILL.md:31`

资源类型	声明权限	推断权限	状态	证据
文件系统	`WRITE`	`WRITE`	✓ 一致	learning-loop.js:8-14 uses fs/promises (readFile, writeFile, mkdir, appendFile)
网络访问	`NONE`	`NONE`	—	No network modules imported or used
命令执行	`NONE`	`NONE`	—	No child_process, exec, spawn, or shell command execution
环境变量	`READ`	`READ`	✓ 一致	learning-loop.js:16-23 reads OPENCLAW_WORKSPACE, LEARNING_* env vars
技能调用	`WRITE`	`WRITE`	✓ 一致	Creates SKILL.md files in ~/.openclaw/skills/<category>/<name>/
剪贴板	`NONE`	`NONE`	—	No clipboard access
浏览器	`NONE`	`NONE`	—	No browser automation
数据库	`NONE`	`NONE`	—	Mentions SQLite FTS5 in docs but no actual database implementation in code

3 项发现

🔗

中危外部 URL 外部 URL

https://hermes-agent.nousresearch.com/

README.md:213

🔗

中危外部 URL 外部 URL

https://agentskills.io/specification

README.md:214

🔗

中危外部 URL 外部 URL

https://www.sqlite.org/fts5.html

SKILL.md:377

目录结构

3 文件 · 33.0 KB · 1150 行

Markdown 2f · 607L JavaScript 1f · 543L

├─ ▾ 📁 scripts

│ └─ 📜 learning-loop.js JavaScript 543L · 16.2 KB

├─ 📝 README.md Markdown 222L · 5.4 KB

└─ 📝 SKILL.md Markdown 385L · 11.5 KB

安全亮点

✓ No shell execution (subprocess/exec/spawn)

✓ No network requests or external IP connections

✓ No credential harvesting (no ~/.ssh, ~/.aws, .env access)

✓ No base64 or obfuscated code execution

✓ No dynamic code evaluation (eval, Function, etc.)

✓ No data exfiltration patterns

✓ No hidden functionality in HTML comments or hidden files

✓ Filesystem operations aligned with documented purpose (skill/memory persistence)