中文 EN

Scan Report

Scan New Files

✓ Trusted — Risk Score 10/100
Last scan:18 hr ago Rescan
10 /100
Claude Agent
作为项目经理操作 Claude Code 完全体。包含：任务执行（提示词设计→执行→监控→质量检查→迭代→汇报）
The claude-agent skill is a well-documented project manager wrapper for Claude Code with transparent shell operations, no credential harvesting, and no data exfiltration beyond its stated notification purpose.
Skill NameClaude Agent
Duration49.8s
Enginepi
✓
Safe to install
No action needed. The skill is safe to use with standard Claude Code permissions.

Findings 1 items

Severity Finding Location
Low
Minor doc-to-code alignment gap Doc Mismatch
SKILL.md states filesystem:NONE but shell scripts technically write to /tmp for PID files and logs. However, these are transient operational files, not project files.
echo $! > "$MONITOR_PID_FILE"
→ Minor cosmetic gap. Consider adding a note about /tmp transient files in SKILL.md, or document filesystem:NONE only for project workspace.
 hooks/start_claude.sh:40
ResourceDeclaredInferredStatusEvidence
Filesystem NONE WRITE ✓ Aligned SKILL.md does not claim filesystem access; Claude Code handles actual file opera…
Network READ READ ✓ Aligned openclaw CLI tool handles notifications; no direct network sockets opened
Shell WRITE WRITE ✓ Aligned tmux + claude subprocess in start_claude.sh:25-26; subprocess.Popen in on_comple…
Environment READ READ ✓ Aligned Reads CLAUDE_AGENT_CHAT_ID, CLAUDE_AGENT_CHANNEL, etc. in on_complete.py:43-68
Clipboard NONE NONE — No clipboard access in any script
Browser NONE NONE — No browser automation in any script
Database NONE NONE — No database access in any script
7 findings
🔗
Medium External URL 外部 URL
https://docs.anthropic.com/en/docs/claude-code
INSTALL.md:8
🔗
Medium External URL 外部 URL
https://docs.anthropic.com/en/docs/claude-code/overview
knowledge/UPDATE_PROTOCOL.md:28
🔗
Medium External URL 外部 URL
https://docs.anthropic.com/en/docs/claude-code/cli-usage
knowledge/UPDATE_PROTOCOL.md:29
🔗
Medium External URL 外部 URL
https://docs.anthropic.com/en/docs/claude-code/settings
knowledge/UPDATE_PROTOCOL.md:30
🔗
Medium External URL 外部 URL
https://docs.anthropic.com/en/docs/claude-code/hooks
knowledge/UPDATE_PROTOCOL.md:31
🔗
Medium External URL 外部 URL
https://docs.anthropic.com/en/docs/claude-code/mcp
knowledge/UPDATE_PROTOCOL.md:32
🔗
Medium External URL 外部 URL
https://docs.anthropic.com/en/docs/claude-code/memory
knowledge/UPDATE_PROTOCOL.md:33

File Tree

22 files · 76.8 KB · 2472 lines
Markdown 15f · 1978L Python 1f · 246L Shell 4f · 246L Text 2f · 2L
├─ ▾ 📁 hooks
│ ├─ 🐍 on_complete.py Python 246L · 7.2 KB
│ ├─ 🔧 pane_monitor.sh Shell 116L · 4.5 KB
│ ├─ 🔧 start_claude.sh Shell 77L · 2.2 KB
│ └─ 🔧 stop_claude.sh Shell 30L · 843 B
├─ ▾ 📁 knowledge
│ ├─ 📝 capabilities.md Markdown 80L · 2.5 KB
│ ├─ 📝 changelog.md Markdown 53L · 2.1 KB
│ ├─ 📝 config_schema.md Markdown 175L · 3.5 KB
│ ├─ 📝 features.md Markdown 145L · 3.7 KB
│ ├─ 📝 prompting_patterns.md Markdown 129L · 2.8 KB
│ └─ 📝 UPDATE_PROTOCOL.md Markdown 57L · 2.0 KB
├─ ▾ 📁 references
│ └─ 📝 claude-code-reference.md Markdown 95L · 2.3 KB
├─ ▾ 📁 scripts
│ └─ 🔧 publish_clawhub.sh Shell 23L · 710 B
├─ ▾ 📁 state
│ ├─ 📄 last_updated.txt Text 1L · 11 B
│ └─ 📄 version.txt Text 1L · 7 B
├─ ▾ 📁 workflows
│ ├─ 📝 knowledge_update.md Markdown 69L · 1.6 KB
│ └─ 📝 standard_task.md Markdown 125L · 2.9 KB
├─ 📝 CHANGELOG.md Markdown 64L · 2.6 KB
├─ 📝 INSTALL.md Markdown 184L · 6.1 KB
├─ 📝 README_EN.md Markdown 119L · 5.2 KB
├─ 📝 README.md Markdown 289L · 11.9 KB
├─ 📝 SECURITY.md Markdown 36L · 1.8 KB
└─ 📝 SKILL.md Markdown 358L · 10.6 KB

Dependencies 4 items

PackageVersionSourceKnown VulnsNotes
bash user-installed system No Shell interpreter required by scripts
tmux user-installed system No Terminal multiplexer required for session management
openclaw user-installed system No CLI tool for notifications and agent wake; used as command executor, not imported
claude user-installed system No Claude Code CLI, the primary target of this wrapper skill

Security Positives

✓ Uses only Python standard library (no external pip dependencies)
✓ SECURITY.md provides comprehensive hardening guidance
✓ Notification mode defaults to 'event' (minimal data disclosure)
✓ No credential harvesting or API key exfiltration
✓ All subprocess calls use the openclaw CLI tool (user-installed)
✓ No base64 encoding, eval(), or obfuscation techniques
✓ No remote script execution (curl|bash, wget|sh)
✓ No access to sensitive paths (~/.ssh, ~/.aws, .env)
✓ Clear separation: skill handles orchestration, Claude Code handles actual file operations
✓ Hardcoded fallback defaults prevent silent failures