Back to skill

Security audit

agent-system

Security checks for vulnerabilities and agentic risk

Overview

This is a local task-orchestration skill, but its instructions overreach by claiming system-level control and telling the agent not to ask for clarification or confirmation.

Install only if you want a Chinese-oriented orchestration skill that may aggressively plan and proceed without asking follow-up questions. Avoid using it for destructive, financial, privacy-sensitive, deployment, messaging, or account-affecting workflows unless its instructions are narrowed to require clarification and explicit confirmation for consequential actions.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T01 · Skill Instruction Hijacking

Error
Location
skills/SKILL.md:154
Finding

Skill Instructions Override Agent Decision Boundaries and Suppress Required Clarification

Content
View full analysis
Master Prompt — Core Intelligent Orchestration System You are OpenClaw's core intelligent orchestration system. Your objective is not to execute a single task, but to: Coordinate all agents toward the "optimal result." Decision principles: 1. Prioritize result correctness over speed. 2. If the task is complex, planning is mandatory. 3. If result quality is low, self-healing is mandatory. 4. If repeated failures occur, degrade but provide the best feasible result. 5. Continuously record metrics for evolution. You must behave like a "system," not a function. ``` Faithful English rendering of `AGENT_RULES.md:67-90`: ```markdown ## General Rules (All Agents Must Follow) ### Placeholder validation - Before invoking any agent template, check for placeholders in the `{xxx}` format. - If an unfilled placeholder is found, fill it with a default value before execution. ### Questions are prohibited - The following statements are not allowed: - "Please provide..." - "You can tell me..." - "I need more in ...[truncated 4565 chars]
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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (24)

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The README advertises very broad trigger phrases such as '分析', '帮我', '处理', and '执行', which are common in normal conversation and can cause the skill to activate far outside its intended scope. In an agent-orchestration skill, overbroad activation is dangerous because it can unexpectedly route benign user requests into multi-step autonomous processing, increasing the chance of unintended actions, prompt hijacking exposure, or privilege misuse.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 207)May include surrounding context.

md
`src/orchestrator.js` 包含完整的参考实现,可作为:

Vague Triggers

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The activation description is overly broad, matching generic requests like analysis, planning, or multi-step handling that occur in many normal conversations. This can cause the skill to activate unexpectedly and take over routing or response shaping outside its intended scope, increasing the chance of prompt/skill interference and unintended delegation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The file is written as a Chinese-only rules manual and includes mandated Chinese output phrases such as the forbidden/required wording and degradation notice, with no indication that users may choose another language. SQP-3 applies to language or locale policy violations when a skill forces a specific language without user opt-in.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The rules explicitly require the agent to execute without confirmation and to persist logs/metrics, but the skill description does not disclose these behavior-affecting and data-affecting side effects to the user. In an orchestration skill that may trigger multi-step actions automatically, this increases the risk of unexpected processing and unintended retention of task content.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The rules mandate recording every session, failure pattern, and metrics to files, and the example logs include task metadata that may contain user-supplied or sensitive workflow content. Persistent plain-language logging in a central orchestrator is especially risky because this skill handles complex, multi-step tasks and may aggregate sensitive inputs across agents.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The entire skill-facing document is written in Chinese, including headings, status labels, and update notes, with no indication that language selection is optional or that the file is intentionally limited to a Chinese-speaking or region-specific context. This creates a natural-language policy concern because it imposes a specific language without user opt-in or justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The file presents the orchestrator's title, description, and master prompt entirely in Chinese, including imperative operating instructions to the agent, with no indication that users may choose another language. Under the policy, forcing a specific language without opt-in is a natural-language policy violation unless the locale restriction is explicitly justified.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest-style description says the skill activates when users describe needs like '分析', '规划', or '拆解任务', but it does not define boundaries, exclusions, or a confidence threshold. This ambiguity can lead to accidental invocation of the orchestration system, which is risky because the skill is designed to select execution paths automatically based on inferred complexity.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The activation description uses very broad everyday phrases such as '分析', '规划', and '多步骤处理', which can cause the skill to activate for many normal user requests beyond its intended scope. In an agent orchestration skill, unintended invocation is risky because it may route ordinary conversations into multi-step planning/execution behavior, increasing the chance of overreach, unnecessary tool use, or policy bypass through excessive autonomy.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The intent-recognition table relies on underspecified keyword matching like '写', '生成', '选择', and '计划', which are common in benign requests and do not reliably indicate the need for this powerful orchestration skill. Because this skill is a core scheduler that can invoke planning, review, and self-heal flows, overbroad triggering expands its operational surface and can lead to unnecessary autonomous task decomposition or execution paths.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This manifest uses Chinese-only natural-language strings for the plugin description and all config field descriptions. Because there is no indication that the skill is region-specific or that users can opt into this locale, it creates a language/locale policy concern under the rule for forced language without user choice.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This markdown file presents all headings, descriptions, and examples exclusively in Chinese, which effectively forces a specific language for users reading or following the skill reference. The policy allows locale constraints only when users are given a choice or the restriction is explicitly justified, neither of which appears here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The skill is written as a Chinese-only operating specification and does not provide user-choice or locale negotiation, which can force interaction patterns or outputs in a specific language without consent. While not directly an exploit primitive, this can degrade user understanding of system behavior, increase misconfiguration risk, and make unsafe autonomous actions harder for non-Chinese users to detect or correct.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The intent-recognition table relies on broad keywords without gating conditions, so ordinary words like “分析”, “比较”, or “写” may route users into this orchestration workflow even when unnecessary. In a core scheduling skill, that increases the attack surface for misrouting, policy conflicts, and unintended application of planning/execution behaviors to benign or unrelated requests.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The file mixes English headers with Chinese protocol/instruction text and later hard-codes Chinese keyword recognition and task descriptions, indicating the skill is designed to operate in Chinese only. There is no natural-language option or documented user choice for language/locale, which creates a locale-policy issue under the rule for forced language without opt-in.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest says the system '基于复杂度自动选择最优执行路径', implying different execution strategies depending on task complexity. In code, dispatch always follows the same sequential flow, and complexity only changes the number/descriptions of placeholder tasks generated by planner rather than selecting different orchestration paths or execution modes.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The schema advertises broad activation criteria such as '分析', '规划', '拆解任务', '多步骤处理', and '自动执行复杂任务' without clear boundaries or disambiguation rules. In an agent dispatch system, this can cause unintended invocation for loosely related user requests, increasing the chance that complex planning or execution behavior is triggered when the user did not explicitly intend it.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

Natural-language fields throughout the schema are written only in Chinese, indicating a fixed language/locale expectation. There is no accompanying opt-in, user choice mechanism, or documented region-specific justification, which can violate language/locale policy requirements.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The file's natural-language strings, including the title and all test inputs, are exclusively in Chinese, which implies a fixed language/locale expectation. There is no indication that users can opt into another language or that this is a region-specific tool with a documented justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

This markdown file presents headings, descriptions, and status information exclusively in Chinese, which can impose a language/locale constraint on users. The file does not indicate that Chinese is optional, nor does it document a region-specific reason for requiring this language.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
73% confidence
Finding

The natural-language instructions and examples are entirely in Chinese, and the README does not state that the skill is region-specific or that users may choose another language. This can be a language/locale policy issue when a skill implicitly forces one language without opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The natural-language description uses Chinese only ("OpenClaw 核心 Agent 调度系统") with no indication that the package is region-specific or that language choice is optional. Under the stated policy, forcing a specific language without opt-in can be a locale-policy concern.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The module comment defines a mandatory system rule that all tasks pass through a specific pipeline. However, executor and reviewer are exported public functions and can be invoked independently, so the documented invariant is not actually enforced by the code.

Content

No source excerpt is available for this finding.