Back to skill

Security audit

调控员

Security checks for vulnerabilities and agentic risk

Overview

This skill is not overtly malicious, but it broadly changes how the agent works, delegates task context to reusable sub-agents, and requires persistent cross-session files without tight scoping.

Install only if you specifically want a broad multi-agent dispatcher that writes persistent task state and shares task context with reusable sub-agent sessions. Avoid using it for sensitive projects unless you can constrain where it writes, disable or review CURSOR_SYNC and memory updates, and ensure delegated sessions do not retain unrelated private context.

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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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 (3)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:25
Finding

Mandatory Workflow and Tool-Control Hijacking

Content
View full analysis
Remediation
View remediation

T02 · Agent Memory Poisoning

Warning
Location
SKILL.md:126
Finding

Persistent Memory and Cross-Session State Manipulation

Content
View full analysis
[brief description] 2. Three violations of the same type trigger a rule-upgrade notification to the user. ``` The following is a faithful English rendering of the complete required closing-write segment at lines 175-182: ```markdown ## Required Closing Set All completed tasks must perform these actions without exception: 1. **Progress file**: complete task-status record. 2. **memory/YYYY-MM-DD.md**: append at least one line to the daily record. 3. **CURSOR_SYNC.md**: update cross-session synchronization between Cursor and OpenClaw. The set is mandatory. The verifier checks it, and a task is not considered complete if any item is missing. ``` ### Technical Analysis The Skill requires persistent writes to progress files, daily memory, a regression log, and a cross-application synchronization file. It also instructs future dispatcher instances to load mutable progress data as the authoritative source after a restart. This design creates a persistent influence channel. Data written during one task can affect later sessions, and the re ...[truncated 1625 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:64
Finding

Excessive Workspace Access and Disclosure to Reusable Sub-Agent Sessions

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (5)

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The file explicitly forbids the dispatcher from using read/write/edit-style tools, then later mandates that it create, read, append, and update multiple files itself. That contradiction can lead to unsafe implementations where guardrails are bypassed or operators incorrectly assume the dispatcher has no direct execution capability when it actually does.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill markets the dispatcher as 'pure scheduling' while its own workflow requires direct local state handling such as creating, reading, and updating progress and sync files. This mismatch is dangerous because it weakens operator trust and can cause the dispatcher to exceed its declared privilege boundary, making auditing and policy enforcement harder.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The description is written as a Chinese-only operational specification and presents the dispatcher behavior in a fixed locale without offering the user any language or locale choice. Per policy, skills should not force a specific language unless the locale restriction is explicitly justified or user-selected.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The activation criteria use open-ended thresholds like '预计 >5 分钟', '步骤 ≥3', and broad task types such as '执行/排障/改配置', which can overlap with many ordinary requests and do not clearly define boundaries. The file also lacks negative examples or clearer scope constraints to show when similar tasks should not invoke this dispatcher flow.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The natural-language description is presented bilingually in Chinese and English, which imposes language expectations on users without stating whether language selection is optional. Under the policy rule, locale or language constraints should either be user-selectable or clearly documented as justified.

Content

No source excerpt is available for this finding.