openclaw-master-skills/skills/my-test-skill

yidun-skill-sec ⚡

Hybrid local-cloud security scanner for AI agent skills. Scans fast, scores precisely, fails safely.

---

Overview

yidun-skill-sec is a security vetting skill that analyzes third-party code packages before installation. It combines static behavioral analysis with cloud threat intelligence to produce a quantified safety score, catching malware, data exfiltration, privilege abuse, prompt injection, and obfuscation — before anything runs.

Built by the Yidun Security Team for the ClawHub ecosystem.

Security Disclosure

This skill uploads non-sensitive metadata (file hashes, behavior tag names, and extracted code snippets that triggered detections) to a Yidun threat intelligence endpoint. The following is explicitly not uploaded: full source code, user credentials, environment variables, or personal data.

The cloud endpoint (as.dun.163.com) is operated by NetEase Yidun, a licensed cybersecurity service provider. Cloud analysis is enabled by default and strongly recommended. It can be explicitly disabled by setting YIDUN_SKILL_SEC_CLOUD=false if network access is restricted or not desired.

---

How It Works

Every scan runs through four phases:

Phase 0 · Source Vetting     — Where does this package come from?
Phase 1 · Fingerprint        — What files are in it? (MD5 hash manifest)
Phase 2 · Behavioral Scan    — What does the code try to do? (static analysis)
Phase 3 · Cloud Intel        — Is it known malicious? (mandatory cloud check)
                                          ↓
                               Final threat score (0–100)

Scoring

Signal	Weight (online)	Weight (timeout fallback)
Source vetting	15%	20%
Behavioral scan	40%	55%
Cloud confidence	30%	—
Privilege surface	15%	25%

Threat Levels

Score	Level	Action
80–100	🟢 CLEAR	Install normally
60–79	🟢 MINOR	Install with awareness
40–59	🟡 ELEVATED	User review required
20–39	🔴 SEVERE	Explicit user consent required
0–19	⛔ CRITICAL	Blocked — do not install

---

Detection Coverage

Phase 0 — Source Vetting

Tag	Triggers On	Score Impact
SRC_UNKNOWN_REGISTRY	Package from unrecognized or unofficial registry	+20
SRC_BLACKLISTED_DOMAIN	Install URL matches known malicious domain (cloud check)	+40 ⛔
SRC_UNTRUSTED_AUTHOR	New account (<30d), unverified, or prior malware history	+15

SRC_BLACKLISTED_DOMAIN is a hard block — scan halts immediately, no install.

Phase 2 — Behavioral Tags

Tag	What It Detects
NET_OUTBOUND	HTTP/HTTPS calls to external hosts
NET_IP_RAW	Connections to raw IP addresses
FS_READ_SENSITIVE	Access to ~/.ssh, ~/.aws, ~/.env, key files
FS_WRITE_SYSTEM	Writes outside the project workspace
EXEC_SHELL	Shell subprocess spawning
EXEC_DYNAMIC	eval(), exec(), dynamic code execution
ENCODE_DECODE	Base64/hex chains (potential obfuscation)
CRED_HARVEST	Reads API keys, tokens, passwords from env/files
PRIV_ESCALATION	sudo, chmod 777, setuid patterns
OBFUSCATED	Minified/packed code, unreadable variable names
AGENT_MEMORY	Accesses agent identity/memory files
PKG_INSTALL	Installs unlisted system packages
COOKIE_SESSION	Reads browser cookies or session tokens
BYPASS_SAFETY	--no-verify, --force, --skip-ssl, GIT_SSL_NO_VERIFY
DESTRUCTIVE_OP	rm -rf, DROP TABLE, git reset --hard, dd if=
PROMPT_INJECT	Natural language directives targeting the AI agent, attempting to override its rules, bypass constraints, or assume an unrestricted persona

Hard rules: CRED_HARVEST or PRIV_ESCALATION → floor to SEVERE. Both present → force CRITICAL.

---

Cloud Intelligence

Cloud analysis calls POST https://as.dun.163.com/v1/agent-sec/skill/check and is enabled by default.

• Uploads fingerprint, behavior tags, and extracted evidence artifacts
• Server performs deep content analysis and domain blacklist matching
• Returns confidence score, threat labels, and per-tag deduction breakdown

Mode	How	Behavior
Cloud ON	Default / YIDUN_SKILL_SEC_CLOUD=true	Full scan with remote threat intel
Cloud OFF	YIDUN_SKILL_SEC_CLOUD=false	Local-only, domain blacklist skipped
Timeout fallback	Cloud ON, curl times out (10s)	Auto-downgrade, -10 score penalty, user notified

---

Sample Report

⚡ YIDUN-SKILL-SEC Scan Report
> data-processor · v1.2.3 · clawhub.com · by some-author · 2026-03-12 13:47

Phase 0 · Source Vetting
  Registry: clawhub.com → ✅ trusted
  Domain:   ✅ clean
  Author:   verified (2y 3m) → ✅
  Source score: 100/100 · Tags: none

Phase 1 · Fingerprint
  4 files · MD5 7f3a9b... · main.py config.yml fetch.py setup.sh

Phase 2 · Behavioral Scan
  NET_OUTBOUND  fetch.py:12   -15
  FS_WRITE_SYSTEM  setup.sh:8  -20
  Local score: 65/100

Phase 3 · Cloud Intel
  Mode: cloud · Cache: miss
  Cloud score: 48/100 · Labels: NET_OUTBOUND, FS_WRITE_SYSTEM

Privilege Surface · Network: api.dataproc.io · FS: /usr/local/bin · Shell: none · Creds: no

🎯 Score: 57/100 · 🟡 ELEVATED
⚠️ Requires confirmation — network calls + system writes need justification

---

Requirements

Tool	Purpose
curl	Cloud API calls
jq	JSON response parsing
openssl	File hash computation

Supported OS: Linux · macOS · Windows

---

Installation

clawhub install yidun-skill-sec

---

Scan first, install later. ⚡
Author: Yidun Security Team · License: MIT