Back to skill

Security audit

Privacy Guard

Security checks for vulnerabilities and agentic risk

Overview

This privacy-scanning skill is not safe to install as published because it bundles private logs containing a token-like secret and writes unredacted sensitive log snippets into local reports.

Do not install this package as-is. The publisher should remove bundled runtime logs, rotate any exposed token-like secret, purge generated report/state files from releases, and change the scanner to redact secrets before writing reports or JSON state.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T09 · Insecure Skill Coding Practices

Error
Location
privacy_guard.py:242
Finding

Unredacted Sensitive Log Content Is Persisted to Local Reports and State Files

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
alert_log.md:163
Finding

Distributed Alert Log Contains a Plaintext API-Token-Like Secret and Private Session Data

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (17)

YARA rule 'backdoor_persistence': Backdoor persistence with malicious payloads (shell commands, SSH key injection, hidden root users) [malware]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · README.md (reported line 193)May include surrounding context.

t stored locally

  • ✅ 不访问任何外部服务器(除非要发送飞书通知)
  • ✅ No external server access (except Feishu notification if configured)

使用场景 / Usage Scenarios

定期巡检 / Regular Scan

bash
# Windows 任务计划 / Windows Task Scheduler
schtasks /create /tn "Privacy-Guard" /tr "python PATH\to\privacy_guard.py --scan" /sc hourly /mo 4

# Linux crontab
0 */4 * * * python /path/to/privacy_guard.py --scan

即时检测 / On-Demand Scan

bash
python privacy_guard.py --scan
python privacy_guard.py --report

更新日志 / Changelog

v0.3.0 (2026-04-11)

  • ✅ 新增三级检测机制 / Three-level detection
  • ✅ 新增容错机制 / Error tolerance mechanism
  • ✅ 新增交互学习 / Interactive learning
  • ✅ 新增自动进化 / Auto evolution

v0.2.0 (2026-04-11)

  • ✅ 收紧检测规则,减少误报 / Tighter rules, fewer false positives
  • ✅ 增加已知安全模式白名单 / Known safe patterns w

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The log reveals handling of sensitive credentials, including what appears to be an API key, without any warning, masking, or safe-handling guidance. Exposure of secrets in logs or conversational context can directly lead to account compromise, unauthorized API usage, and lateral access to other connected systems.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instruction to upload evolving protection-related work to GitHub creates a direct exfiltration path from a local environment containing sensitive operational and financial context. Even if framed as documentation or backup, natural-language summaries and copied local content can leak secrets, private holdings, paths, or internal procedures to a remote repository.

Content

No source excerpt is available for this finding.

Ssd 4

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

This sequence progressively steers the agent toward autonomous skill acquisition and broader authority, which is a classic escalation pattern. In context, the agent is being taught to self-expand capabilities, search for new skills, and apply them later, increasing the attack surface far beyond the original task scope.

Content

No source excerpt is available for this finding.

Ssd 4

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The trust-building language culminates in requests to search for, install, and operationalize more skills for ongoing future use, which is a strong privilege-escalation and persistence signal. In an agent framework, this can convert a one-off interaction into long-term expanded control with network, tool, and filesystem consequences.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The content endorses broad trigger conditions for skill use without clear scope limits, effectively encouraging the agent to activate capabilities in loosely defined situations. That increases the chance of unintended actions, privilege creep, and unsafe execution paths because the agent is being pushed toward autonomous behavior without tight guardrails.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
94% confidence
Finding

Deletion-oriented cleanup instructions are presented without warning about irreversible outcomes, which is unsafe in a system capable of modifying local files. This normalizes destructive actions and increases the likelihood that the agent will perform unsafe deletions based on ambiguous user intent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The natural-language instructions, commands, and descriptions in this skill are exclusively presented in Chinese, with no indication that users may choose another language or that the locale restriction is intentional. Under the policy, forcing a specific language without user opt-in is a locale-policy violation.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The log contains an instruction to run an external package via npx -y @tencent-weixin/openclaw-weixin-cli without pinning an exact version. This creates a supply-chain risk because future package updates or a compromised latest release could change behavior and execute unreviewed code in a privileged local environment.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The content documents instructions to reorganize or delete user files without emphasizing the destructive impact or need for backups and confirmation. In an agent-operated local environment, such vague cleanup actions can cause data loss, break workflows, or delete evidence needed for recovery.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

This is a second occurrence of the same unsafe pattern: invoking an unpinned external package with npx -y trusts the latest upstream artifact at execution time. In an agent context, this is especially dangerous because it can turn a chat instruction into remote code execution through package substitution or malicious updates.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The tool scans for sensitive data in logs, then persists excerpts of those same sensitive lines into scan_report.md and suspicious.json. That creates a secondary data-at-rest exposure: secrets or personal data may remain on disk longer, in a more discoverable location, and potentially with weaker access controls than the original transient logs.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The CLI description, prompts, commands, and user-facing status messages are presented in Chinese only, and the command parser expects Chinese phrases such as 确认安全 and 确认风险. This imposes a specific language requirement on users without any opt-in, fallback, or stated regional justification, which matches the language/locale policy violation criteria.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The documented example --cmd='确认安全 1' and the responses in process_user_command imply full natural-language commands are supported, but main() splits args.cmd and passes only parts[0] as the command. As a result, process_user_command receives 确认安全 instead of the full string it checks with if "确认安全" in command, while commands like 添加白名单 某个模式 are reduced to 添加白名单 and lose their payload, contradicting the advertised behavior.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The interactive help lists commands like safe N, risk N, and add P, suggesting direct support for those actions. However, those branches call process_user_command("确认安全", item_id) / process_user_command("确认风险", item_id), and process_user_command lowercases input and only recognizes Chinese phrases such as 确认安全, 确认风险, 查看可疑, and 添加白名单, creating a mismatch between the displayed user intent and actual command semantics.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The sensitive_types values are written only in Chinese (持仓明细, 金额数据, 账号密码, 个人信息). For a generally scoped config file, this imposes a specific language/locale in natural-language content without any visible opt-in or explanation that the skill is intended only for a Chinese-language or region-specific environment.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

This markdown file contains user-facing natural language entirely in Chinese, including the title, status text, and report labels. Under the policy criteria, forcing a specific language without user opt-in or a documented justification is a locale-policy issue.

Content

No source excerpt is available for this finding.