可信 — 风险评分 5/100
上次扫描:1 天前 重新扫描
5 /100
agent-harness
Long-running agent workflow automation — feature tracking, progress management, and session orchestration
The agent-harness skill is a legitimate project workflow automation tool that manages feature tracking and session orchestration using documented bash scripts and git operations.
技能名称agent-harness
分析耗时31.7s
引擎pi
可以安装
This skill is safe to use. No security concerns identified.
资源类型声明权限推断权限状态证据
文件系统 NONE WRITE ✓ 一致 scripts/harness.sh:51-54 - cp command copies templates
命令执行 NONE WRITE ✓ 一致 scripts/harness.sh - git init, chmod +x operations are documented and necessary
环境变量 NONE READ ✓ 一致 scripts/harness.sh:67 - date command for timestamps only
网络访问 NONE NONE No network calls in codebase
2 项发现
🔗
中危 外部 URL 外部 URL
https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents
README.md:3
🔗
中危 外部 URL 外部 URL
https://docs.anthropic.com/en/docs/about-claude/models
README.md:296

目录结构

12 文件 · 38.3 KB · 1140 行
Markdown 7f · 700L Shell 2f · 311L JSON 2f · 107L Text 1f · 22L
├─ 📁 docs
│ ├─ 📝 architecture.md Markdown 56L · 2.1 KB
│ ├─ 📝 file-flow.md Markdown 63L · 2.0 KB
│ └─ 📝 sequence.md Markdown 73L · 2.4 KB
├─ 📁 prompts
│ ├─ 📝 coder.md Markdown 85L · 2.9 KB
│ └─ 📝 initializer.md Markdown 57L · 2.2 KB
├─ 📁 scripts
│ └─ 🔧 harness.sh Shell 244L · 7.9 KB
├─ 📁 templates
│ ├─ 📄 claude-progress.txt Text 22L · 765 B
│ ├─ 📋 feature_list.json JSON 34L · 784 B
│ └─ 🔧 init.sh Shell 67L · 1.8 KB
├─ 📝 README.md Markdown 296L · 10.2 KB
├─ 📋 skill.json JSON 73L · 2.0 KB
└─ 📝 SKILL.md Markdown 70L · 3.2 KB

依赖分析 3 项

包名版本来源已知漏洞备注
bash >=4.0 system Standard shell dependency
git >=2.0 system Version control dependency
python3 >=3.7 system JSON manipulation only

安全亮点

✓ All shell operations are clearly documented and directly support stated purpose
✓ File operations are scoped to project directory only
✓ No credential harvesting or sensitive data access
✓ No obfuscation techniques detected
✓ No external network calls except to Anthropic documentation URLs
✓ Python3 usage limited to JSON parsing/manipulation
✓ Git operations follow standard version control workflow
✓ No curl/wget remote script execution
✓ init.sh template has all sensitive code commented out