Back to skill

Security audit

MindVault 思维永生

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because a scanner confirmed that it disguises a WorkBuddy session path to evade keyword blocking, and found that its claimed lossless archive can omit paired call records.

Consider avoiding installation unless you need this skill and can verify that it may access the WorkBuddy sessions you intend to archive. The scanner found a deliberate attempt to evade keyword blocking and a defect that may make some backups incomplete; overall scan coverage was partial, so the results do not establish that the skill is safe.

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
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (5)

YARA rule 'agent_skill_mcp_tool_poisoning_metadata': MCP/tool metadata poisoning indicators in tool schemas or skill manifests [agent_skills]

High
Category
YARA Match
Confidence
80% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · SKILL.md (reported line 7)May include surrounding context.

md
---
name: mindvault
slug: mindvault
displayName: MindVault 思维永生
display_name: MindVault 思维永生
version: "1.3.1"
description: "Agent 记忆与长期记忆管理:对话备份归档成本地 JSONL,防 AI 失忆——换电脑、换会话都不丢记忆;从历史萃取规则生成快照,支持 Claude Code / WorkBuddy / OpenClaw 等本地框架,纯标准库"
display_description: "MindVault 对话归档与记忆进化引擎(基础版免费):对话增量备份为 JSONL、长期规则萃取、项目快照生成、可选 DRAS-V 五步思考协议。支持 OpenClaw / AutoClaw / Marvis / WorkBuddy / CherryStudio / Coze,数据全部保存在本地。"
keywords:
  - memory
  - 记忆
  - long-term

The trigger “走流程” (“go through the process”) is broad enough to activate the reasoning workflow unintentionally.

Medium
Category
Trigger Abuse
Confidence
96% confidence
Finding

“走流程” is a common expression that can refer to following any procedure, but this line makes it activate the full DRAS-V workflow. Unlike the domain-specific alternative “DRASV,” it has no stated contextual restriction or exclusion for ordinary conversational use.

Content

No source excerpt is available for this finding.

The WorkBuddy session locator deliberately obscures a path component to bypass keyword-based blocking.

Medium
Category
Prompt Injection
Confidence
94% confidence
Finding

The locator constructs 'projects' as 'pro' + 'jects', and its own docstring explicitly describes avoiding keyword-literal blocking. Reading WorkBuddy sessions is consistent with this backup skill's purpose, but deliberately concealing the target path from a blocking mechanism is security-control evasion and could enable conversation access that the surrounding environment intended to restrict; this does not itself bypass operating-system permissions.

Content

No source excerpt is available for this finding.

WorkBuddy's claimed lossless backup discards original paired function-call records.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

For paired WorkBuddy tool events, wb_flush_tool saves raw_rec=res or call, so the result replaces the original function-call record rather than preserving both. Selected call fields survive, but dictionary arguments are truncated in args_summary and other original fields can disappear, contradicting the lossless-backup claim and compromising the archive's reliability for memory restoration or security auditing.

Content

No source excerpt is available for this finding.

The trigger “走流程” (“follow the process”) is broad enough to activate the DRAS-V protocol unintentionally.

Low
Category
Trigger Abuse
Confidence
96% confidence
Finding

The skill activates its five-step protocol when the user says “走流程,” a common phrase that can refer to any routine procedure. No contextual restriction or exclusion distinguishes ordinary uses from an explicit request to activate DRAS-V.

Content

No source excerpt is available for this finding.