claw-sergeant 安全扫描报告 — 低风险 | ClawSafe

25 /100

claw-sergeant

Train autonomous OpenClaw AI agents through LLM-guided curriculum design and multi-turn dialogue evaluation

The skill is a legitimate AI agent training framework with a missing required dependency that would cause runtime failure, plus minor documentation hygiene issues.

技能名称claw-sergeant

分析耗时44.8s

引擎pi

✓

可以安装

Fix the missing learning_logger.py file before deployment. The skill performs declared LLM-based training with CLI interaction for a documented purpose.

安全发现 2 项

严重性	安全发现	位置
中危	Missing required module causes runtime failure 供应链 learning_logger.py is imported by main.py and trainer.py but the file does not exist in the codebase. This will cause an ImportError when the skill is executed. `from learning_logger import LearningLogger` → Create the missing learning_logger.py module or remove the import if logging functionality is not needed.	`main.py:15`
低危	Incomplete training insights file 文档欺骗 TRAINING_INSIGHTS.md contains placeholder HTML content with a broken asset reference (logo.png), suggesting incomplete or boilerplate documentation. `<img src='assets/logo.png'>` → Replace placeholder content with actual training insights documentation.	`.claw_sergeant_accumulated_lessons/TRAINING_INSIGHTS.md:4`

资源类型	声明权限	推断权限	状态	证据
文件系统	`WRITE`	`WRITE`	✓ 一致	main.py:73 - writes training_results.json
网络访问	`READ`	`READ`	✓ 一致	llm_handler.py:72 - POSTs to LLM API endpoint
命令执行	`WRITE`	`WRITE`	✓ 一致	claw_agent.py:44 - executes openclaw CLI command
环境变量	`READ`	`READ`	✓ 一致	main.py:91-95 - reads LLM_API_KEY, CLAW_RECIPIENT

4 项发现

🔗

中危外部 URL 外部 URL

https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control

.gitignore:107

🔗

中危外部 URL 外部 URL

https://pdm-project.org/en/latest/usage/project/#working-with-version-control

.gitignore:114

🔗

中危外部 URL 外部 URL

https://abstra.io/docs

.gitignore:181

🔗

中危外部 URL 外部 URL

https://docs.cursor.com/context/ignore-files

.gitignore:200

目录结构

14 文件 · 57.8 KB · 1739 行

Python 7f · 1330L Ignore 1f · 213L Markdown 5f · 193L Text 1f · 3L

├─ ▾ 📁 .claw_sergeant_accumulated_lessons

│ ├─ 📝 CURRICULUM_PATTERNS.md Markdown 5L · 113 B

│ ├─ 📝 ERRORS.md Markdown 5L · 108 B

│ └─ 📝 TRAINING_INSIGHTS.md Markdown 5L · 115 B

├─ 📄 .gitignore Ignore 213L · 4.7 KB

├─ 🐍 claw_agent.py Python 76L · 2.3 KB

├─ 🐍 config.py Python 33L · 1.3 KB

├─ 🐍 curriculum.py Python 169L · 5.0 KB

├─ 🐍 llm_handler.py Python 133L · 4.2 KB

├─ 🐍 main.py Python 205L · 6.7 KB

├─ 📝 README.md Markdown 4L · 107 B

├─ 📄 requirements.txt Text 3L · 43 B

├─ 📝 SKILL.md Markdown 174L · 8.0 KB

├─ 🐍 test_phases.py Python 226L · 7.7 KB

└─ 🐍 trainer.py Python 488L · 17.5 KB

依赖分析 4 项

包名	版本	来源	已知漏洞	备注
`httpx`	`>=0.27`	pip	否	Version specifier present
`loguru`	`>=0.7`	pip	否	Version specifier present
`python-dotenv`	`>=1.0`	pip	否	Version specifier present
`openclaw`	`unknown`	external CLI	否	Must be installed separately - not a Python dependency

安全亮点

✓ All core functionality (LLM API calls, CLI execution) is declared in SKILL.md

✓ No base64-encoded payloads or obfuscated code detected

✓ No network exfiltration or credential theft beyond necessary API authentication

✓ No suspicious file accesses (no ~/.ssh, ~/.aws, or .env file reading)

✓ Shell execution is limited to documented openclaw CLI tool

✓ Training results written only to declared local files

✓ httpx library used with proper timeout configuration (120s)