Back to skill

Security audit

zettaranc-perspective

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious because scanners report broad activation triggers and instructions for persistent financial preferences and guardrail changes; intent is not established, and scan coverage is incomplete.

Consider the suspicious verdict and incomplete evidence before installing. Review whether you are comfortable with the reported cross-session retention of financial preferences, automatic updates to persistent guidance, and broad activation triggers; the scan does not establish malicious intent.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
Findings (16)

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · INSTALL.md (reported line 69)May include surrounding context.

B2. 配置数据源

bash
cp .env.example .env

编辑 .env,三选一:

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · knowledge/harness.md (reported line 8)May include surrounding context.

md
>
> 本文件从 SKILL.md v3.3.0 抽出。原位置:SKILL.md 第 824-1322 行(v3.3.1 拆分)。

<!-- Skill-Runtime
加载时机: Agent 需要对照约束规则、验证输出、纠错恢复时
用途: 完整的 Guardrails / Feedback Loop / Error Recovery / Context Management 规范
大小: ~22KB

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · knowledge/harness.md (reported line 8)May include surrounding context.

md
>
> 本文件从 SKILL.md v3.3.0 抽出。原位置:SKILL.md 第 824-1322 行(v3.3.1 拆分)。

<!-- Skill-Runtime
加载时机: Agent 需要对照约束规则、验证输出、纠错恢复时
用途: 完整的 Guardrails / Feedback Loop / Error Recovery / Context Management 规范
大小: ~22KB

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · INSTALL.md (reported line 35)May include surrounding context.

md
# 2. 链接到 Claude Code 的 skills 目录
mkdir -p ~/.claude/skills
ln -s "$(pwd)/zettaranc-perspective" ~/.claude/skills/zettaranc-perspective

# 或者直接复制(不跟随更新)
# cp -R zettaranc-perspective ~/.claude/skills/

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · INSTALL.md (reported line 89)May include surrounding context.

dotenv
DATA_MODE=jnb
TUSHARE_TOKEN=你的token
TUSHARE_API_URL=https://api.waditu.com/dataapi

注意 TUSHARE_API_URL 必须是路径式中转地址,官方根地址 https://api.tushare.pro 不行。

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · INSTALL.md (reported line 147)May include surrounding context.

bash
mkdir -p ~/.claude/skills
ln -s /path/to/zettaranc-perspective ~/.claude/skills/zettaranc-perspective

重启 Claude Code。现在问「帮我看看 600519.SH」应走真实数据;问「什么是四块砖」应走知识库。

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
85% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · INSTALL.md (reported line 201)May include surrounding context.

Skill 侧:

bash
ls -l ~/.claude/skills/zettaranc-perspective/SKILL.md

然后在 Claude Code 里问「什么是少妇战法」和「帮我看看 600519」各一次:前者应引用知识库,后者应调用 zt analyze。

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · INSTALL.md (reported line 201)May include surrounding context.

Skill 侧:

bash
ls -l ~/.claude/skills/zettaranc-perspective/SKILL.md

然后在 Claude Code 里问「什么是少妇战法」和「帮我看看 600519」各一次:前者应引用知识库,后者应调用 zt analyze。

File System Enumeration

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code scans file system directories looking for sensitive files. This could be reconnaissance for credential theft.

Content

Scanner excerpt · INSTALL.md (reported line 201)May include surrounding context.

Skill 侧:

bash
ls -l ~/.claude/skills/zettaranc-perspective/SKILL.md

然后在 Claude Code 里问「什么是少妇战法」和「帮我看看 600519」各一次:前者应引用知识库,后者应调用 zt analyze。

Cross-session retention of financial preferences lacks a user-facing privacy notice.

Medium
Category
Output Handling
Confidence
87% confidence
Finding

The instructions call for cross-session retention of watchlists, trading style, and risk preferences without explaining storage, retention duration, or deletion controls. Voluntarily discussing financial preferences is not equivalent to consenting to persistent profiling, and durable storage could expose sensitive financial interests beyond the original conversation.

Content

No source excerpt is available for this finding.

The self-improvement workflow changes persistent guardrails without requiring user notification.

Medium
Category
Output Handling
Confidence
83% confidence
Finding

The workflow directs review results to update the knowledge base and guardrails without requiring approval or separating recommendations from durable instruction changes. If implemented by a write-capable agent, incorrect or manipulated performance records could persistently bias later financial guidance; the examples concern trading rules, not demonstrated modification of platform security controls.

Content

No source excerpt is available for this finding.

The loading condition includes broad topics such as health and ability without clearly requiring a request for this life-decision framework.

Medium
Category
Trigger Abuse
Confidence
91% confidence
Finding

L006 activates the module for questions about life direction, health, ability, temperament, and fortune. Although framed as life decisions, these topics are broad enough to encompass ordinary health or skills questions, and no exclusions distinguish those requests from requests for the four-circle framework.

Content

No source excerpt is available for this finding.

Trade-review triggers include everyday phrases without a required trading context.

Medium
Category
Trigger Abuse
Confidence
95% confidence
Finding

L221 lists “复盘” (review/retrospective) and “交作业” (submit homework) as triggers, although both commonly occur outside financial discussions. The JNB-mode restriction controls feature availability, but these trigger rules do not explicitly require stock-trading intent or exclude ordinary retrospectives and schoolwork.

Content

No source excerpt is available for this finding.

Generic terms such as cycles and negative feedback can activate the financial macro module outside its intended domain.

Low
Category
Trigger Abuse
Confidence
90% confidence
Finding

L006 lists cycles, macro topics, contrarian operations, four-year cycles, and negative feedback as loading conditions without explicitly requiring a financial-market context. Several terms also occur in engineering, science, and everyday discussion, creating ambiguity about when this investment module should load.

Content

No source excerpt is available for this finding.

The firewall trigger does not distinguish trading risk checks from cybersecurity firewalls.

Low
Category
Trigger Abuse
Confidence
87% confidence
Finding

L006 includes the unqualified term “防火墙” (firewall) among the module's loading conditions. The body uses that metaphor for a three-layer trading check, but the trigger itself could also match ordinary cybersecurity questions.

Content

No source excerpt is available for this finding.

The manifest describes broad career, life, and business decision coverage without defining when this particular perspective should be invoked.

Low
Category
Trigger Abuse
Confidence
83% confidence
Finding

L04 identifies the persona and financial specialties but also includes career, life, and business decisions without invocation constraints. Those broad areas overlap with many general assistance requests, and the manifest does not distinguish an explicit request for this perspective from an ordinary decision-support question.

Content

No source excerpt is available for this finding.