Back to skill

Security audit

xqant daily gushouplus report

Security checks for vulnerabilities and agentic risk

Overview

The skill is mainly a financial report workflow, but it should be reviewed because it references a persistent every-60-second cron job that auto-posts to an unspecified chat channel and also directs access to global memory.

Install only after verifying the cron job ID, payload, destination channel, and deletion controls. Disable automatic posting unless the channel audience is known and approved, remove the MEMORY.md dependency if the packaged product list is sufficient, and confirm Wind API quota impact before enabling scheduled runs.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • System PersistenceInstalls backdoors, hooks, services, or scheduled tasks that survive the run
  • 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
Findings (2)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:25
Finding

Unnecessary Access to Global Agent Memory

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 25–29
Vulnerability Type: Least-privilege violation through unnecessary persistent-memory access
Risk Level: Medium

Vulnerable Code Snippet:

markdown
### 1. 读取产品清单
从以下权威源读取 87 只产品代码:
- **Primary**: `STANDARD/蚂蚁固收 + 竞品清单 (最终版).md`
- **Secondary**: `MEMORY.md`
- **本地**: `references/product_codes.md`

The same global memory source is listed again at lines 125–129:

markdown
## 参考文件

- **产品清单**: `STANDARD/蚂蚁固收 + 竞品清单 (最终版).md`
- **执行规范**: `STANDARD/固收 + 竞品监控执行规范.md`
- **MEMORY**: `MEMORY.md`
- **产品代码**: `references/product_codes.md`

Technical Analysis

The Skill directs the Agent to read the global MEMORY.md file while performing a product-list lookup. Access to persistent Agent memory is not required for this task because the package already provides the complete product list in references/product_codes.md.

This violates least-privilege principles by expanding the Skill's data-access scope beyond the information needed to generate the report. If MEMORY.md contains private conversation-derived state, user preferences, credentials, internal instructions, or unrelated operational context, that content could enter the Skill's active context.

The reviewed files do not direct the Agent to transmit memory contents externally, so no confirmed exfiltration behavior was identified. The vulnerability is the unnecessary access boundary itself.

Attack Path

  1. A user activates the Skill by requesting a daily fixed-income product report.
  2. The Agent follows the documented product-source workflow.
  3. The Agent reads the global MEMORY.md because it is designated as a secondary authoritative source.
  4. Unrelated persistent information becomes available in the Skill's active context.
  5. That information may influence subsequent analysis or inadvertently appear in generated output.

Impact Assessment

The Skill ...[truncated 437 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove both references to MEMORY.md from SKILL.md.
  2. Use only the packaged references/product_codes.md as the product-list source.
  3. If an external list must be supported, require the user to provide a specific, narrowly scoped file explicitly.
  4. Define an allowlist of project-relative files the Skill may read.
  5. Prevent automatic access to global memory, conversation archives, home-directory files, or unrelated workspace content.
  6. Validate the local list's count, format, and uniqueness without consulting persistent Agent state.

T06 · System Persistence

Error
Location
SKILL.md:134
Finding

Opaque High-Frequency Persistent Cron Job with Automatic Publication

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 134–148
Vulnerability Type: Underspecified persistent scheduled execution exceeding the declared task scope
Risk Level: High

Vulnerable Code Snippet:

markdown
## Cron 任务集成

已创建 cron 任务 (ID: `8b2d83f7-0444-42c9-87a3-e37b577c52c7`):
- **名称**: 固收 + 87 只产品全量查询
- **频率**: 每 60 秒自动触发
- **模式**: isolated session + agentTurn payload
- **输出**: 自动发布到聊天频道

使用方法:
```bash
# 手动触发
cron run --jobId 8b2d83f7-0444-42c9-87a3-e37b577c52c7 --runMode force

# 查看执行历史
cron runs --jobId 8b2d83f7-0444-42c9-87a3-e37b577c52c7
text

### Technical Analysis

The Skill advertises an already-created cron job that runs every 60 seconds in an isolated session and automatically publishes output to a chat channel. This behavior materially exceeds the stated purpose of producing one report at 21:30 on each trading day.

The job is opaque: the repository does not include its creation command, complete `agentTurn` payload, publication destination, authorization model, consent controls, or removal procedure. Consequently, a reviewer cannot verify what instructions execute under the referenced job ID or where the resulting information is published.

The repository does not contain code that creates the scheduled task, so installation of persistence by the packaged files was not directly confirmed. The confirmed issue is that the Skill instructs users or Agents to rely on and invoke an external, persistent, high-frequency job without sufficient provenance or lifecycle controls.

### Attack Path

1. A user or Agent trusts the cron integration described by the Skill.
2. The documented `cron run` command is executed against the hard-coded job ID.
3. The pre-existing job executes an undisclosed `agentTurn` payload in an isolated session.
4. Its output is automatically published to an unspecified chat channel.
5. Independently of manual execution, the documented schedule con
...[truncated 933 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the hard-coded reference to the opaque pre-existing job.
  2. Replace the 60-second schedule with an explicit once-daily 21:30 schedule limited to applicable trading days.
  3. Include the complete scheduled payload in the reviewed repository.
  4. Display the exact publication destination and require explicit user approval before creating or enabling the schedule.
  5. Provide transparent commands for creation, inspection, pausing, and permanent deletion.
  6. Apply least-privilege credentials and narrowly scoped publication permissions.
  7. Add execution limits, API quotas, timeout handling, duplicate-publication prevention, and failure backoff.
  8. Require confirmation before forced manual execution.
  9. Log each execution and publication destination without recording secrets or unrelated Agent context.
  10. Ensure uninstalling or disabling the Skill also removes any schedule it created.
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 (5)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger conditions are broad natural-language phrases such as '21:30 产品复盘' and '87 只产品数据', which can cause the skill to activate on loosely related user requests. In this skill's context, accidental activation can launch large-scale data queries and downstream automated reporting behavior, increasing the risk of unintended processing and publication.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill states that cron job output is '自动发布到聊天频道' without a clear disclosure, consent, or audience restriction mechanism. In this context, the skill aggregates and publishes financial product analysis at regular intervals, so unintended disclosure to a shared channel could expose sensitive internal research, create compliance issues, or distribute inaccurate partial results automatically.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The title and opening description are entirely in Chinese and present the skill reference as the required interface/documentation, with no indication that another language is supported or that the user can choose their preferred language. This can violate language/locale policy when a skill imposes a specific language without explicit opt-in or justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This file contains user-facing natural-language instructions only in Chinese, including the title, usage guidance, and output descriptions. Under the stated policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale restriction is clearly justified, which is not documented here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
70% confidence
Finding

整个技能标题、用途、流程和示例输出均固定为中文表述,且未声明是否支持其他语言或允许按用户偏好切换。按照语言/区域策略,这种默认单一语言约束在缺少用户选择或正当业务说明时可能构成自然语言策略问题。

Content

No source excerpt is available for this finding.