Back to skill

Security audit

私域专家团 · 马甲实战版

Security checks for vulnerabilities and agentic risk

Overview

The skill’s marketing advice, customer records, and requested updates fit its documented purpose, with disclosed privacy and installation considerations.

Choose redacted storage and review reports before sharing; local records are unencrypted. Specify the customer explicitly when restoring or reporting, because defaults depend on the working directory. Drafting can reuse that customer's saved context unless you opt out. Requested updates download and run an unpinned installer and change globally available project skills, so review the update source first.

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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (30)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared purpose describes advisory workflows and factual verification, not software maintenance or global skill installation. The supplied commands perform only an update-and-install workflow, changing local code and global skill availability. These are substantive capabilities not disclosed by the description, rather than routine implementation details of answering private-domain operations questions.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied chunk performs software/skill installation rather than the declared consulting or fact-verification functions. Although the repository name relates to the declared subject, globally installing its skills is a distinct environment-changing operation not disclosed in the description. The chunk does not establish what the installed skills subsequently do.

Content

No source excerpt is available for this finding.

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · tools/siyu_team/eval/static.py (reported line 156)May include surrounding context.

python
def _effective_rule_id(rule: LexiconRule, token: str) -> str:
    if rule.flag != "COMPLIANCE_RED":
        return rule.rule
    return {
        "guarantee": "deceptive_guarantee",
        "restricted_tool": "restricted_tool",

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · tools/siyu_team/eval/static.py (reported line 197)May include surrounding context.

python
"""Return ``(flag, description, severity, hard)`` for one occurrence."""

    if rule.flag in {"NO_CALIBRATION", "NO_METRIC", "PRIVACY_COLLECT"}:
        return rule.flag, rule.description, rule.severity, False

    reason = _context_reason(text, start, end, mode)
    if reason is not None:

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · tools/siyu_team/eval/static.py (reported line 236)May include surrounding context.

python
description = rule.description
        if not hard:
            description += f";{mode.value} 语境仅提示或转人工复核"
        return rule.flag, description, rule.severity, hard

    if rule.flag == "INDUCE_SHARE":
        if mode is ScanMode.CUSTOMER_COPY:

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · tools/siyu_team/eval/static.py (reported line 240)May include surrounding context.

python
if rule.flag == "INDUCE_SHARE":
        if mode is ScanMode.CUSTOMER_COPY:
            return rule.flag, rule.description, rule.severity, True
        hard = mode is ScanMode.KNOWLEDGE and _is_prescriptive(
            text, start, end
        )

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · tools/siyu_team/eval/static.py (reported line 247)May include surrounding context.

python
description = rule.description
        if not hard:
            description += f";{mode.value} 语境保留风险提示,不阻断"
        return rule.flag, description, rule.severity, hard

    if rule.flag == "COMPLIANCE_RED":
        kind = _red_match_kind(token)

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · tools/siyu_team/eval/static.py (reported line 252)May include surrounding context.

python
if rule.flag == "COMPLIANCE_RED":
        kind = _red_match_kind(token)
        if mode is ScanMode.CUSTOMER_COPY:
            return rule.flag, rule.description, rule.severity, True
        if kind == "superlative":
            hard = (
                mode is ScanMode.KNOWLEDGE

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · tools/siyu_team/eval/static.py (reported line 274)May include surrounding context.

python
description = rule.description
        if not hard:
            description += f";{mode.value} 语境保留风险提示,不阻断"
        return rule.flag, description, rule.severity, hard

    return rule.flag, rule.description, rule.severity, rule.default_hard

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · tools/siyu_team/eval/static.py (reported line 276)May include surrounding context.

python
description += f";{mode.value} 语境保留风险提示,不阻断"
        return rule.flag, description, rule.severity, hard

    return rule.flag, rule.description, rule.severity, rule.default_hard


def _match_hit(

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_read.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The skill requires Chinese output without user opt-in.

Medium
Category
Output Handling
Confidence
99% confidence
Finding

The instruction “中文、具体动词” explicitly requires responses in Chinese rather than merely requesting clear language. The file provides neither a user language choice nor a documented regional requirement justifying this restriction.

Content

No source excerpt is available for this finding.

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 · modules/_expert-team/compliance-critic.md (reported line 25)May include surrounding context.

md
7. **对照马甲的判断**:团长交给你的马甲判断,和你的结论冲突时,说清是这家条件不同还是判断要改。

## 你绑定的 skills
- `skills/wechat-compliance-redlines/SKILL.md`

## 输出契约
按五段输出:现状盘点 / 核心问题 / 可落地动作 / 最脆弱的前提 / 合规风险。给可落地细节,不要泛泛而谈、不灌鸡汤金句。

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 · modules/_expert-team/content-product-officer.md (reported line 25)May include surrounding context.

md
7. **对照马甲的判断**:团长交给你的马甲判断,和你的结论冲突时,说清是这家条件不同还是判断要改。

## 你绑定的 skills
- `skills/content-as-product/SKILL.md`
- `skills/reactivation-playbook/SKILL.md`

## 输出契约

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 · modules/_expert-team/content-product-officer.md (reported line 26)May include surrounding context.

md
## 你绑定的 skills
- `skills/content-as-product/SKILL.md`
- `skills/reactivation-playbook/SKILL.md`

## 输出契约
按五段输出:现状盘点 / 核心问题 / 可落地动作 / 最脆弱的前提 / 合规风险。给可落地细节,不要泛泛而谈、不灌鸡汤金句。

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 · modules/_expert-team/ops-ad-officer.md (reported line 25)May include surrounding context.

md
7. **对照马甲的判断**:团长交给你的马甲判断,和你的结论冲突时,说清是这家条件不同还是判断要改。

## 你绑定的 skills
- `skills/ops-as-ad-funnel/SKILL.md`
- `skills/conversion-caliber/SKILL.md`

## 输出契约

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 · modules/_expert-team/ops-ad-officer.md (reported line 26)May include surrounding context.

md
## 你绑定的 skills
- `skills/ops-as-ad-funnel/SKILL.md`
- `skills/conversion-caliber/SKILL.md`

## 输出契约
按五段输出:现状盘点 / 核心问题 / 可落地动作 / 最脆弱的前提 / 合规风险。给可落地细节,不要泛泛而谈、不灌鸡汤金句。

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 · modules/_expert-team/private-pr-officer.md (reported line 25)May include surrounding context.

md
7. **对照马甲的判断**:团长交给你的马甲判断,和你的结论冲突时,说清是这家条件不同还是判断要改。

## 你绑定的 skills
- `skills/trust-asset-playbook/SKILL.md`
- `skills/crisis-response-sop/SKILL.md`

## 输出契约

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 · modules/_expert-team/private-pr-officer.md (reported line 26)May include surrounding context.

md
## 你绑定的 skills
- `skills/trust-asset-playbook/SKILL.md`
- `skills/crisis-response-sop/SKILL.md`

## 输出契约
按五段输出:现状盘点 / 核心问题 / 可落地动作 / 最脆弱的前提 / 合规风险。给可落地细节,不要泛泛而谈、不灌鸡汤金句。

The module provides metric-definition guidance that the parent manifest explicitly delegates to majia-huiyuan.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

L18–L23 instruct the skill to define conversion-rate numerators, denominators, reporting periods, and cumulative-versus-period metrics directly. The majia-siyu manifest explicitly excludes metric-caliber deliverables and requires questions involving “怎么算、口径” to be routed to majia-huiyuan, so this module crosses that stated boundary.

Content

No source excerpt is available for this finding.

Everyday phrases are listed as activation triggers without explicit exclusions.

Medium
Category
Trigger Abuse
Confidence
98% confidence
Finding

The triggers include 「打包」 (package), 「整理一份」 (organize a copy), and 「给老板或客户看」 (show the boss or client). These phrases commonly occur outside client-report generation, and the skill does not explicitly restrict their activation to requests to merge stored client archives.

Content

No source excerpt is available for this finding.

The skill mandates Chinese output without user opt-in.

Medium
Category
Output Handling
Confidence
99% confidence
Finding

「一律中文」 explicitly requires Chinese in all cases, without offering a language choice or respecting the user's requested language. No region-specific compliance requirement is documented to justify this blanket restriction.

Content

No source excerpt is available for this finding.

Everyday phrases activate archive restoration without clear scope restrictions.

Medium
Category
Trigger Abuse
Confidence
98% confidence
Finding

Broad conversational triggers such as “last time” and “continue” can initiate reading and summarizing local customer archives without establishing that the user means a previous customer consultation. Because the default customer can be inferred from the current directory, an ambiguous request could expose an unintended customer's business information in the conversation. Path containment and confirmation when multiple alternative customers exist reduce risk, but do not resolve ambiguity for the initially selected customer.

Content

No source excerpt is available for this finding.

The skill mandates Chinese output without user opt-in.

Medium
Category
Output Handling
Confidence
100% confidence
Finding

「一律中文」 explicitly requires Chinese for all output, regardless of the user's language preference. The skill provides neither a language choice nor a documented regional requirement justifying this restriction.

Content

No source excerpt is available for this finding.

The skill requires all user-facing output to be in Chinese without user opt-in.

Medium
Category
Output Handling
Confidence
99% confidence
Finding

The instruction “面向用户一律中文” unconditionally requires Chinese for user-facing communication. The skill offers no language choice and provides no region-specific justification for overriding the user's preferred language.

Content

No source excerpt is available for this finding.