Back to skill

Security audit

诉讼律师全能助手(专业版·不限国内)

Security checks for vulnerabilities and agentic risk

Overview

The skill is marked suspicious because scanners report undeclared local log inspection, automatic retrieval and execution of an external helper, and workflows that may expose confidential legal reports; the evidence is incomplete.

Given the suspicious verdict, avoid installing this skill for confidential legal work unless you can review and trust the external helper and service, understand how case files and reports are handled, and disable or carefully review the log-inspection and report-sharing workflows. The scan was partial: some referenced local files were unavailable, so it could not fully verify those parts.

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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (8)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied command performs conversation-log inspection rather than any declared legal research, calculation, drafting, or verification function. No connection is shown that would make this log access a supporting implementation detail of the legal assistant. The snippet does not show transmission of the extracted identifier or establish what the invoked Python script does internally.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

Installing a runner is reasonably related to the legal-assistance purpose, so the description mismatch alone is not the vulnerability. However, the workflow automatically clones or updates an unpinned external repository and subsequently executes its Python scripts with access to API credentials and confidential case materials; compromise of either upstream source could therefore become local code execution and data theft.

Content

No source excerpt is available for this finding.

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · SKILL.md (reported line 171)May include surrounding context.

md
ROOT=$(python3 ~/.cue/cue-skills/cue-research/scripts/cue_api.py root)
K=$(python3 -c "import json;print(json.load(open('$HOME/.cue/config.json'))['api_key'])" 2>/dev/null || echo "$CUE_API_KEY")
echo "=== 1/4 API Key ==="; [ -n "$K" ] && echo "已配置(${#K} 位)" || echo "未配置!"
echo "=== 2/4 Cue 服务 ==="; curl -sS --max-time 10 "https://cuecue.cn/api/health" | python3 -c "import sys,json;print('status =',json.load(sys.stdin).get('status'))" 2>/dev/null || echo "服务不可达(网络/代理/沙箱出站)"
echo "=== 3/4 搭子在线情况 ==="; curl -sS --max-time 15 "https://cuecue.cn/api/playbook" -H "Authorization: Bearer $K" | python3 -c "
import sys,json
want={'template_M3loak':'法律文书审校','template_BFSwVa':'合同风控与合规审查','template_LwdgVW':'疑难法律实操案例库','template_c84pmj':'诉讼文书起草','template_sRdpsI':'法律实务问题研判','template_uJKQlK':'国内法规调研','template_UsZH8W':'法规逐条深度解读','template_jbjaEU':'中外法律对比'}

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · SKILL.md (reported line 172)May include surrounding context.

md
K=$(python3 -c "import json;print(json.load(open('$HOME/.cue/config.json'))['api_key'])" 2>/dev/null || echo "$CUE_API_KEY")
echo "=== 1/4 API Key ==="; [ -n "$K" ] && echo "已配置(${#K} 位)" || echo "未配置!"
echo "=== 2/4 Cue 服务 ==="; curl -sS --max-time 10 "https://cuecue.cn/api/health" | python3 -c "import sys,json;print('status =',json.load(sys.stdin).get('status'))" 2>/dev/null || echo "服务不可达(网络/代理/沙箱出站)"
echo "=== 3/4 搭子在线情况 ==="; curl -sS --max-time 15 "https://cuecue.cn/api/playbook" -H "Authorization: Bearer $K" | python3 -c "
import sys,json
want={'template_M3loak':'法律文书审校','template_BFSwVa':'合同风控与合规审查','template_LwdgVW':'疑难法律实操案例库','template_c84pmj':'诉讼文书起草','template_sRdpsI':'法律实务问题研判','template_uJKQlK':'国内法规调研','template_UsZH8W':'法规逐条深度解读','template_jbjaEU':'中外法律对比'}
sc=json.load(sys.stdin).get('data',{}).get('scenes',[])

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · SKILL.md (reported line 179)May include surrounding context.

md
live={b.get('template_id') for s in sc for b in s.get('buddies',[])}
for t,n in want.items(): print(('  ✅ ' if t in live else '  ⚠️ '),n,t,'' if t in live else '(playbook 未列出)')
"
echo "=== 4/4 剩余积分 ==="; curl -sS --max-time 10 "https://cuecue.cn/api/v1/billing/subscriptions/current" -H "Authorization: Bearer $K" | python3 -c "
import sys,json
d=json.load(sys.stdin) or {}
d=d.get('data',d) or {}

The categorical no-retention assurance contradicts the workflow's server-side report recovery and later retrieval.

High
Category
Excessive Agency
Confidence
94% confidence
Finding

The skill instructs the assistant to give categorical safety and no-retention assurances while uploading complete case files to an external service and supporting replay and later report retrieval. These features do not prove that source files are retained, but they make the blanket assurance unsupported and potentially misleading, which could induce lawyers to disclose privileged information without an informed assessment of retention and access controls.

Content

No source excerpt is available for this finding.

The workflow mandates promotional referrals and incentivized social sharing of litigation reports, beyond legal-assistance tasks.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The skill makes promotional referrals, purchase offers, and incentivized report sharing mandatory parts of the assistant's workflow rather than optional account help. This redirects the assistant from the user's legal task toward the provider's commercial objectives, and encouraging report sharing is particularly risky where outputs may contain privileged or sensitive litigation information; however, the text does not instruct automatic purchases or publication.

Content

No source excerpt is available for this finding.

Report-sharing workflow omits a warning about publicly exposing confidential case information.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The workflow encourages sharing a report to a public social platform for credits and passes the actual report identifier into that process without requiring a confidentiality check or redaction. Litigation reports may contain client identities, evidence, and legal strategy, so following this incentive could cause serious disclosure; the supplied text does not establish that merely opening the link publishes the report or that report links are unauthenticated.

Content

No source excerpt is available for this finding.