Back to skill

Security audit

Agent Collaboration Framework

Security checks for vulnerabilities and agentic risk

Overview

This skill is a coherent multi-agent collaboration guide, but it asks for broad session visibility and cross-workspace file handoffs that users should review before installing.

Install only if you intentionally want agents to see and coordinate across sessions and workspaces. Before use, narrow session visibility where possible, require confirmation before spawning sessions or writing HANDOFF files, avoid sensitive data in handoffs, and prefer archiving or logging handoff receipt instead of deleting the only copy.

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

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:119
Finding
Global Session Visibility Violates Least-Privilege Isolation## Vulnerability Details **File Location**: `SKILL.md`, lines 119–127 **Vulnerability Type**: Excessive session access permissions **Risk Level**: Medium The documented configuration enables global session visibility: ```markdown ## 安裝配置 ### 1. openclaw.json — sessions 可見性 ```json { "tools": { "sessions": { "visibility": "all" } } } ``` ``` ### Technical Analysis Setting `tools.sessions.visibility` to `"all"` grants session visibility beyond the department-specific collaboration boundaries described elsewhere in the skill. This configuration is broader than necessary for agents that only need access to sessions associated with their assigned tasks. If an untrusted, compromised, or incorrectly configured agent has access to the session tools, global visibility may allow it to discover or interact with unrelated sessions. This weakens isolation between departments and conflicts with the skill's stated scope-isolated operating model. The issue is an access-control weakness rather than evidence of intentional malicious behavior. Exploitation depends on the affected agent also possessing session-tool capabilities. ### Attack Path 1. An administrator applies the documented configuration and sets session visibility to `"all"`. 2. A low-privilege departmental agent is granted access to session discovery or messaging tools. 3. The agent becomes compromised, receives malicious instructions, or acts outside its intended scope. 4. The agent uses globally visible session metadata to identify sessions belonging to other departments. 5. Where supported by its tool permissions, the agent reads session context or sends messages into those unrelated sessions. 6. Information from finance, operations, content, or other departments may be disclosed or manipulated outside the intended trust boundary. ### Impact Assessment A successful exploit could permit cross-department session discovery, unauthorized disclosure of conversation context, and unintended ...[truncated 495 chars]
Remediation
## Remediation Suggestions 1. Replace global session visibility with explicit per-agent or per-department access controls. 2. Allow each agent to access only sessions it created, sessions explicitly delegated to it, or sessions associated with an approved collaboration. 3. Use narrowly scoped allowlists for authorized agent and session identifiers. 4. Separate session discovery, read, and send permissions so agents receive only the operations required for their duties. 5. Revoke temporary cross-department access immediately after a collaboration ends. 6. Validate authorization server-side for every session read or message operation rather than relying solely on visibility settings. 7. Record and monitor cross-department session access, including the requesting agent, target session, operation, and authorization decision. 8. Document any exceptional need for global visibility and restrict that capability to a trusted coordinating agent such as Main.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Vague Triggers

Medium
Confidence
97% confidence
Finding
The skill advertises very broad trigger phrases such as '跨部门', '协同', and '让XX做', which are common in normal user conversation and can cause the skill to activate in situations not specifically intended for multi-agent orchestration. Because this skill enables task dispatch and cross-agent collaboration, accidental activation could route user requests into a more powerful workflow than expected, increasing the chance of unintended tool use, delegation, or data exposure across agent boundaries.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The HANDOFF mechanism explicitly instructs agents to write files into other workspaces and delete files after reading, but it does not require user awareness, confirmation, audit logging, or safety checks around file paths and contents. In a multi-agent environment this creates a real integrity and confidentiality risk: agents may modify local state, pass sensitive data between scopes, or erase evidence of the exchange without clear user consent.

Static analysis

No suspicious patterns detected.