SkillLite Rust Binary

A lightweight Skills secure execution engine for Linux and macOS.

This package is part of a Cargo Workspace. Build from the repository root.

Build & Install Reference

All commands run from repository root (or use cd skilllite && cargo install --path .).

Build commands

Install commands

Default features = sandbox, audit, agent. memory_vector is optional.

Output paths

• Install: ~/.cargo/bin/skilllite or ~/.cargo/bin/skilllite-sandbox
• Build: target/release/skilllite or target/release/skilllite-sandbox

Project Structure (Workspace)

skilllite/           # Repository root
├── Cargo.toml       # [workspace] members = ["skilllite", "crates/*"]
├── skilllite/       # This dir: main binary entry
│   └── src/         # main.rs, cli, commands, stdio_rpc, mcp
└── crates/
    ├── skilllite-core/      # Config, skill, path_validation, observability
    ├── skilllite-sandbox/   # Sandbox + env (independently addable via cargo add)
    ├── skilllite-executor/  # Session, transcript, memory
    └── skilllite-agent/     # LLM, tools, chat, extensions

crates/ follows the common Rust multi-crate layout (e.g. tokio, clap, serde), enabling:

• Enforced dependency direction and no circular deps
• Independent publishing (e.g. skilllite-sandbox via cargo add)
• Faster incremental builds

Features

• SKILL.md Parsing: Parse skill metadata from YAML front matter with auto-detection of entry points and language
• Dependency Management: Automatic Python venv and Node.js node_modules setup with caching
• Multi-level Sandbox Execution:
  • Level 1: No sandbox - direct execution
  • Level 2: Sandbox isolation only (macOS Seatbelt / Linux namespace + seccomp)
  • Level 3: Sandbox isolation + static code scanning (default)
• Security Scanning: Static code analysis with customizable rules for Python and JavaScript
• Network Control: Configurable network access derived from compatibility field
• Resource Limits: Memory limits (default 256MB) and timeout controls (default 30s)
• JSON Protocol: stdin/stdout JSON communication

Installation

cargo install --path .

CLI Commands

Skill Management

Skill Execution

Tools & Adapters

Agent & Chat

Other Commands

Usage Details

Run a Skill

skilllite run <SKILL_DIR> '<INPUT_JSON>' [OPTIONS]

Options:

• --allow-network: Override SKILL.md network policy
• --cache-dir <DIR>: Custom cache directory for environments
• --max-memory <MB>: Maximum memory limit in MB (default: 256)
• --timeout <SECS>: Execution timeout in seconds (default: 30)
• --sandbox-level <1|2|3>: Sandbox security level (default: 3)

Example:

skilllite run ./examples/python_skill '{"message": "Hello, World!"}'

Execute a Script Directly

skilllite exec <SKILL_DIR> <SCRIPT_PATH> '<INPUT_JSON>' [OPTIONS]

Execute a specific script without requiring an entry_point in SKILL.md.

Options:

• --args <ARGS>: Script arguments
• --allow-network: Allow network access
• --sandbox-level <1|2|3>: Sandbox security level

Scan a Skill

skilllite scan <SKILL_DIR> [--preview-lines <N>]

List all executable scripts in a skill directory (JSON output for LLM analysis).

Validate a Skill

skilllite validate <SKILL_DIR>

Show Skill Info

skilllite info <SKILL_DIR>

Security Scan a Script

skilllite security-scan <SCRIPT_PATH> [OPTIONS]

Options:

• --allow-network: Allow network operations
• --allow-file-ops: Allow file operations
• --allow-process-exec: Allow process execution

SKILL.md Format

SkillLite follows the Claude Agent Skills Specification.

---
name: my-skill
description: My awesome skill
compatibility: "Requires Python 3.x, network access"
license: MIT
metadata:
  author: Your Name
  version: "1.0.0"
---

# My Skill

Description of the skill...

Key Fields:

• name: Skill name (max 64 chars, lowercase + hyphens)
• description: What the skill does (max 1024 chars)
• compatibility: Environment requirements (auto-detects language and network needs)
• license: License name or reference
• metadata: Additional metadata (author, version, etc.)

Auto-detection:

• Entry point is auto-detected from scripts/main.{py,js,ts,sh}
• Language is detected from compatibility field or entry point extension
• Network access is enabled if compatibility mentions "network", "internet", "http", "api", or "web"

Environment Variables

Full env var list: ENV_REFERENCE / ENV_REFERENCE 中文.

Custom Security Rules

Create a .skilllite-rules.yaml file in your skill directory:

use_default_rules: true
disabled_rules:
  - py-file-open
rules:
  - id: custom-rule
    pattern: "dangerous_function\\s*\\("
    issue_type: code_injection
    severity: high
    description: "Custom dangerous function"
    languages: ["python"]

Python Integration

import subprocess
import json

def run_skill(skill_dir: str, input_data: dict) -> dict:
    result = subprocess.run(
        ["skilllite", "run", skill_dir, json.dumps(input_data)],
        capture_output=True,
        text=True
    )
    if result.returncode != 0:
        raise RuntimeError(result.stderr)
    return json.loads(result.stdout)

# Usage
output = run_skill("./my_skill", {"param": "value"})

License

MIT