skill-autoinstaller（技能自动安装程序）

Skill Auto-Installer

Automatically discover, evaluate, validate, and install new PopeBot skills from GitHub repositories.

Quick Start

# Discover skills from a GitHub repo
node /job/.pi/skills/skill-autoinstaller/discover.js --repo zeroclaw-labs/zeroclaw

# Evaluate a specific skill
node /job/.pi/skills/skill-autoinstaller/evaluate.js --path /tmp/skills/awesome-skill

# Install a skill
node /job/.pi/skills/skill-autoinstaller/install.js --path /tmp/skills/awesome-skill

# Full pipeline: discover → evaluate → install
node /job/.pi/skills/skill-autoinstaller/pipeline.js --repo openclaw/openclaw --pattern "skills/*"

Commands

discover.js

Scans a GitHub repository for potential skills.

# Scan by repo
node /job/.pi/skills/skill-autoinstaller/discover.js --repo zeroclaw-labs/zeroclaw

# Scan local directory
node /job/.pi/skills/skill-autoinstaller/discover.js --dir /path/to/repo

# With search pattern
node /job/.pi/skills/skill-autoinstaller/discover.js --repo openclaw/openclaw --pattern "skills/*"

# JSON output
node /job/.pi/skills/skill-autoinstaller/discover.js --repo zeroclaw-labs/zeroclaw --json

evaluate.js

Performs security audit and compatibility check on a skill directory.

# Evaluate skill
node /job/.pi/skills/skill-autoinstaller/evaluate.js --path /tmp/skills/my-skill

# With security audit
node /job/.pi/skills/skill-autoinstaller/evaluate.js --path /tmp/skills/my-skill --security

# JSON output
node /job/.pi/skills/skill-autoinstaller/evaluate.js --path /tmp/skills/my-skill --json

install.js

Installs a validated skill into PopeBot.

# Install skill
node /job/.pi/skills/skill-autoinstaller/install.js --path /tmp/skills/my-skill

# With activation
node /job/.pi/skills/skill-autoinstaller/install.js --path /tmp/skills/my-skill --activate

# Dry run
node /job/.pi/skills/skill-autoinstaller/install.js --path /tmp/skills/my-skill --dry-run

pipeline.js

Runs the full discovery → evaluation → installation pipeline.

# Full pipeline
node /job/.pi/skills/skill-autoinstaller/pipeline.js --repo zeroclaw-labs/zeroclaw

# With pattern matching
node /job/.pi/skills/skill-autoinstaller/pipeline.js --repo openclaw/openclaw --pattern "skills/*"

# Interactive mode
node /job/.pi/skills/skill-autoinstaller/pipeline.js --repo zeroclaw-labs/zeroclaw --interactive

Security Audit Checks

The evaluator performs these security checks (inspired by ZeroClaw's audit system):

1. Manifest Validation: SKILL.md must exist with valid frontmatter
2. Path Traversal: No .. in paths or symlinks outside sandbox
3. Dangerous Patterns: No eval, exec, child_process without sandboxing
4. Secret Handling: No hardcoded credentials
5. Network Calls: External calls must be intentional and documented
6. File Operations: Write operations must be confined to /job/tmp/ or skill directory
7. Permission Model: Skills declare required permissions in metadata

Skill Metadata Format

Skills should include this metadata in SKILL.md frontmatter:

---
name: skill-name
description: What the skill does
homepage: https://github.com/author/skill
metadata:
  {
    "popebot":
      {
        "emoji": "🔧",
        "requires": { "bins": ["node", "git"] },
        "permissions": ["network", "filesystem"],
        "install":
          [
            {
              "id": "npm",
              "kind": "npm",
              "package": "@scope/package",
              "bins": ["tool"],
              "label": "Install tool (npm)",
            },
          ],
      },
  }
---

Installation Strategies

The auto-installer supports multiple installation methods:

1. npm: npm install -g package
2. brew: brew install formula
3. apt: apt-get install package
4. pip: pip install package
5. go: go install package@latest
6. binary: Download from release URL
7. script: Run installation script
8. source: Build from source

Error Handling

• Failed discovery: Returns empty list with error message
• Failed evaluation: Skill marked as "incompatible" with reasons
• Failed installation: Rollback to previous state, detailed error log

Output Formats

All commands support --json for machine-readable output:

{
  "status": "success",
  "skills": [
    {
      "name": "skill-name",
      "path": "/tmp/skills/skill-name",
      "evaluation": { "passed": true, "score": 95 },
      "installed": true
    }
  ]
}