Back to skill

Security audit

FollowinMCP

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent Followin market-intelligence package, but it documents unsafe API-key placement and can automatically use a saved watchlist in remote requests.

Review this before installing. Use Followin API keys only through headers or a protected secret store, not in SSE URLs. Disable or require confirmation for saved watchlist use unless you are comfortable sending those tickers to Followin, and be aware that broad market or Telegram wording may route requests into specialized remote data workflows.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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)

T09 · Insecure Skill Coding Practices

Warning
Location
references/01-followin-intel-center.SKILL.md:20
Finding

API Key Exposed in SSE URL Query Parameters

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
references/10-macro-morning-brief.SKILL.md:83
Finding

Implicit Access to Persistent User Watchlist and Transmission to a Remote MCP Service

Content
View full analysis
Remediation
View remediation
“Would you like me to include your saved watchlist? Its ticker symbols will be sent to the Followin MCP service for price retrieval.” 3. Display the exact ticker symbols before transmission and allow the user to remove sensitive entries. 4. Clearly identify the external destination and explain what data will be sent. 5. Provide a privacy-preserving mode that excludes all personalized watchlist processing. 6. Apply data minimization by transmitting only tickers necessary for the current request. 7. Avoid logging watchlist parameters locally and request appropriate retention controls from the MCP provider. 8. Document persistent-memory access and external data handling in the Skill metadata so hosts can enforce permission prompts. 9. Where supported, separate memory-reading permission from network-tool permission and require authorization for both operations. ]]>
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (35)

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill explicitly routes vague prompts like '市场怎么样' and all 'KOL喊单' style queries to this skill by default, which creates an overly broad catch-all entry point. In an agent setting, this can cause misrouting of unrelated financial requests into a trading-signal workflow, increasing the chance of inappropriate tool use, overcollection of market data, or bypass of narrower domain-specific skills and guardrails.

Content

No source excerpt is available for this finding.

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 · references/08-btc-macro-dashboard.SKILL.md (reported line 56)May include surrounding context.

md
---

<!-- Original slash-command frontmatter (reference):
name: BTC Macro Dashboard
description: 评估BTC当前宏观环境,输出0-100综合评分和分层分析。当用户问"BTC宏观怎么样"、"宏观环境如何"、"现在几分"时触发。
trigger: BTC宏观、BTC宏观看盘、BTC宏观评分、BTC宏观环境、BTC macro、BTC macro dashboard、BTC macro score、BTC macro environment、how is BTC macro

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 · references/08-btc-macro-dashboard.SKILL.md (reported line 56)May include surrounding context.

md
---

<!-- Original slash-command frontmatter (reference):
name: BTC Macro Dashboard
description: 评估BTC当前宏观环境,输出0-100综合评分和分层分析。当用户问"BTC宏观怎么样"、"宏观环境如何"、"现在几分"时触发。
trigger: BTC宏观、BTC宏观看盘、BTC宏观评分、BTC宏观环境、BTC macro、BTC macro dashboard、BTC macro score、BTC macro environment、how is BTC macro

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 · references/09-gold-macro-dashboard.SKILL.md (reported line 55)May include surrounding context.

md
---

<!-- Original slash-command frontmatter (reference):
name: Gold Macro Dashboard
description: 评估黄金当前宏观环境,输出0-100综合评分和分层分析。当用户问"黄金宏观怎么样"、"黄金环境如何"、"黄金现在几分"时触发。
trigger: 黄金宏观、黄金宏观看盘、黄金宏观评分、黄金宏观环境、gold macro、gold macro dashboard、gold macro score、gold macro environment、how is gold macro

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 · references/09-gold-macro-dashboard.SKILL.md (reported line 55)May include surrounding context.

md
---

<!-- Original slash-command frontmatter (reference):
name: Gold Macro Dashboard
description: 评估黄金当前宏观环境,输出0-100综合评分和分层分析。当用户问"黄金宏观怎么样"、"黄金环境如何"、"黄金现在几分"时触发。
trigger: 黄金宏观、黄金宏观看盘、黄金宏观评分、黄金宏观环境、gold macro、gold macro dashboard、gold macro score、gold macro environment、how is gold macro

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 · references/11-us-stock-earnings-report.SKILL.md (reported line 55)May include surrounding context.

md
---

<!-- Original slash-command frontmatter (reference):
name: US Stock Earnings Report
description: 单股财报三维分析(财务Beat/Miss + 媒体情绪 + 宏观背景)。必须指定具体股票代码或公司名才触发,如"帮我看AAPL财报"、"TSLA earnings"、"英伟达财报分析"。泛问"今天有哪些财报"走情报中心,不在本Skill范围。
trigger: 帮我看XX财报、XX财报分析、XX财报速查、XX earnings、XX earnings report、[股票代码]财报、[公司名]财报、[ticker] earnings、earnings report、earnings analysis、show me [ticker] earnings、look at [ticker] earnings

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 · references/11-us-stock-earnings-report.SKILL.md (reported line 55)May include surrounding context.

md
---

<!-- Original slash-command frontmatter (reference):
name: US Stock Earnings Report
description: 单股财报三维分析(财务Beat/Miss + 媒体情绪 + 宏观背景)。必须指定具体股票代码或公司名才触发,如"帮我看AAPL财报"、"TSLA earnings"、"英伟达财报分析"。泛问"今天有哪些财报"走情报中心,不在本Skill范围。
trigger: 帮我看XX财报、XX财报分析、XX财报速查、XX earnings、XX earnings report、[股票代码]财报、[公司名]财报、[ticker] earnings、earnings report、earnings analysis、show me [ticker] earnings、look at [ticker] earnings

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 · references/12-macro-analyzer.SKILL.md (reported line 52)May include surrounding context.

md
---

<!-- Original slash-command frontmatter (reference):
name: Macro Analyzer
description: 宏观数据影响分析 — 从指标变动到板块影响的全链路。必须同时包含"指标名"+"影响/解读/分析"才触发,如"CPI出来了对市场有什么影响"、"非农解读"、"利率影响哪些板块"。纯数据查询("CPI是多少")走情报中心,综合日报走morning-brief,BTC/黄金宏观走各自看盘Skill。
trigger: 宏观指标影响、宏观指标解读、宏观指标分析、宏观数据影响、宏观数据解读、宏观数据分析、CPI影响、CPI解读、CPI分析、非农影响、非农解读、非农分析、利率影响、利率解读、GDP影响、GDP解读、关税影响、关税分析、macro impact、macro analysis、CPI impact、NFP impact、rate impact、GDP impact、tariff impact、indicator analysis

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 · references/12-macro-analyzer.SKILL.md (reported line 52)May include surrounding context.

md
---

<!-- Original slash-command frontmatter (reference):
name: Macro Analyzer
description: 宏观数据影响分析 — 从指标变动到板块影响的全链路。必须同时包含"指标名"+"影响/解读/分析"才触发,如"CPI出来了对市场有什么影响"、"非农解读"、"利率影响哪些板块"。纯数据查询("CPI是多少")走情报中心,综合日报走morning-brief,BTC/黄金宏观走各自看盘Skill。
trigger: 宏观指标影响、宏观指标解读、宏观指标分析、宏观数据影响、宏观数据解读、宏观数据分析、CPI影响、CPI解读、CPI分析、非农影响、非农解读、非农分析、利率影响、利率解读、GDP影响、GDP解读、关税影响、关税分析、macro impact、macro analysis、CPI impact、NFP impact、rate impact、GDP impact、tariff impact、indicator analysis

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file’s top-level description is entirely Chinese and does not indicate that other languages are supported. Per the policy, forcing a specific language or locale without user choice is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The operational instructions for the agent are only provided in Chinese, which can effectively constrain usage to a single language without explicit opt-in or justification. The file does not state that Chinese is required for a region-specific purpose.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill instructs users to place the API key directly in the SSE URL query string while also duplicating it in a header. Query parameters are commonly exposed through logs, browser/history storage, reverse proxies, monitoring systems, and error reports, so this pattern increases the chance of credential leakage beyond the intended recipient.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly instructs users to embed API keys in both the SSE URL query string and request headers. Putting secrets in URLs increases the chance they will be exposed through logs, browser history, proxy/server telemetry, crash reports, screenshots, or copied configuration files; duplicating the key in headers further broadens accidental disclosure surface. In this skill context, the risk is more credible because the file is an onboarding/configuration document that users are likely to copy verbatim into MCP host settings.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly routes ambiguous requests that omit a clear domain qualifier to the crypto daily brief by default. This can cause unintended skill activation and produce responses from the wrong capability, which is a security-relevant routing flaw because user intent may be misinterpreted and the agent may access or prioritize inappropriate tools or data sources.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill contains contradictory routing rules for KOL-related queries: earlier sections say queries like 'ETH的KOL观点' should be handled by this skill, while the confusion table routes 'KOL怎么看BTC' and '大V怎么看ETH' to a different skill. This inconsistency can cause misrouting of user requests, producing incorrect tool invocation and potentially exposing users to the wrong class of outputs, especially if a strategy-oriented skill returns trading-oriented content instead of neutral information aggregation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The trigger keywords are broad terms such as '策略', '大户', 'KOL', '仓位', and '能不能做', which are common in many adjacent market-analysis contexts. Without tighter activation constraints at the trigger point, the skill may capture requests that belong to news, macro, funding-rate, or research workflows, leading to confused routing and potentially unsafe or misleading downstream analysis.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill explicitly states that any user query mentioning "TG" or "Telegram" must be routed here, even when the request may actually belong to another specialized skill. This can cause scope hijacking and misrouting, leading the agent to fetch or summarize the wrong data source, miss safer or more appropriate workflows, and degrade trust or policy compliance in multi-skill environments.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The description says the skill triggers on phrases such as “宏观环境如何” and “现在几分”, which are generic expressions that could be used in many unrelated conversations. Although one example is BTC-specific, the listed trigger set is not consistently constrained to BTC, increasing the risk of unintended invocation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The skill’s stated purpose is narrow BTC macro scoring, but it explicitly instructs the agent to perform direct HTTP and Web retrieval outside the MCP-scoped toolset. That expands the trust boundary, increases prompt-injection and data-exfiltration surface from arbitrary web content, and undermines least-privilege expectations for a scoring skill.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

General Web search/fetch capability is broader than necessary for this use case and allows the skill to ingest arbitrary untrusted web content when answering a narrowly defined BTC macro question. In an agent setting, this materially raises the risk of prompt injection, misleading data sourcing, and unintended retrieval from attacker-controlled pages.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The intent rules say inputs like “宏观看盘”, “宏观环境”, and “今天几分” activate the skill, but these phrases are broad and can refer to non-BTC topics. The section does not consistently specify exclusion conditions for general market or unrelated scoring requests.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The activation text says the skill triggers when users ask phrases like "黄金环境如何" and "黄金现在几分", which are relatively broad and could overlap with ordinary conversation about gold without clearly distinguishing this skill from other gold-analysis tasks. The active frontmatter description does not include explicit negative examples or tighter scope constraints, so invocation boundaries are ambiguous.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The documentation instructs users to place the API key both in the SSE URL query string and in headers. Query-string secrets are commonly exposed through logs, browser/history artifacts, proxy telemetry, configuration screenshots, and error reporting, which increases the chance of credential leakage even if the endpoint itself is legitimate.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill instructs operators to place real API keys directly into SSE endpoint URLs and headers for network-connected services without any warning about safe secret handling. Embedding secrets in URLs is dangerous because they may be exposed through logs, browser history, proxy telemetry, crash reports, or host configuration files, increasing the risk of credential leakage and unauthorized API use.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Earlier sections instruct using finance_tool_economic_calendar directly with no parameters and explicitly say stable_request with economic_calendar is invalid, while the report template at L233 and notes at L273 describe the calendar as coming via stable_request or an economic_calendar path with from/to parameters. The same section also labels treasury and dollar-index sources as FRED in the output template even though earlier steps say those FRED calls were removed in favor of finance_tool_treasury_rates and FMP DXUSD, creating active documentation-to-instructions contradictions.

Content

No source excerpt is available for this finding.