Back to skill

Security audit

analyst-fundamentals

Security checks for vulnerabilities and agentic risk

Overview

This is a real A-share stock analysis workflow, but it asks for broad scraping, local file writes, subagents, and actionable trading-style recommendations that are not fully disclosed or aligned with its stated no-advice purpose.

Review this skill carefully before installing. Use it only in a workspace where creating many analysis files is acceptable, validate stock codes to a strict six-digit format before running, expect it to contact Eastmoney, and treat any generated recommendation or position-sizing language as informational rather than personalized financial advice.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (3)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:97
Finding

Unvalidated Stock-Code Interpolation Enables Command Injection and Path Traversal

Content
View full analysis
` directory. The same unvalidated placeholder is subsequently reused throughout the project for source files, generated analyses, and the final report. Consequently, a single malicious value can affect many filesystem operations. ### Attack Path 1. An attacker submits a value presented as a stock code but containing shell metacharacters, path separators, or traversal components. 2. The Agent substitutes the value into `mkdir -p data/{code}` or the absolute workspace path. 3. If the command is passed to a shell, the injected syntax can execute an additional command with the Agent process's privileges. 4. If the value is passed to filesystem APIs, traversal can redirect the data directory outside the expect ...[truncated 675 chars]
Remediation
View remediation

T01 · Skill Instruction Hijacking

Warning
Location
phase1_guide.md:126
Finding

Scraped Web Content Is Passed to Subagents Without Prompt-Injection Isolation

Content
View full analysis
Remediation
View remediation

other

Warning
Location
prompts/c1_three_questions.md:236
Finding

Declared Neutral-Analysis Scope Conflicts With Trading and Position-Sizing Instructions

Content
View full analysis
take a heavy position. Two "YES" answers -> take a light position. One "YES" answer -> avoid. Zero "YES" answers -> exclude. ``` `prompts/c1_three_questions.md:236-263`, translated faithfully: ```text | Number of YES answers | Recommendation | Action | | 3 YES | Strongly recommended | Heavy position permitted | | 2 YES + 1 MAYBE | Recommended | Position may be opened | | 2 YES + 1 NO | Cautiously recommended | Try a light position | | 1 YES + 2 MAYBE/NO | Not currently recommended | Observe | | 0-1 YES | Not recommended | Exclude | Example conclusion: The comprehensive judgment is a strong recommendation, and value and growth investors are advised to allocate a heavy position. ``` Related trading-signal language also appears in `prompts/b6_fund_market.md:46-49`, where fund-flow thresholds are labeled strong buy, moderate buy, or sell signals. ### Technical Analysis The top-level Skill declaration tells users that the tool performs only fundamental analysis and does not provide investment-decision advice. Downstream prompts nevertheless direct the Agent to issue recommendations, buy or sell signals, position-entry guidance, and explicit position sizing. This is not merely descriptive financial analysis. Terms such as “open a position,” “heavy position,” “light position,” and “sell signal” are actionable investment instructions. The disclaimer ...[truncated 1288 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (28)

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The guide instructs creation of a user-specific local directory and batch writing of many markdown files, which exceeds a narrow 'analyze fundamentals' function and can overwrite or accumulate local data without clear scoping. Local filesystem access is sensitive because it creates persistence on the host and may affect user data or consume storage unexpectedly.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The workflow plans bulk creation of multiple .md files but does not tell the user that local files will be created or potentially overwritten. This is dangerous because users may unknowingly permit persistent writes, lose previous data, or be unable to distinguish generated artifacts from their own files.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The guide instructs automated access to a third-party site to retrieve stock data, but it does not disclose that network requests will be sent to an external service. Undisclosed external access can expose user interest patterns, create compliance issues, and violate user expectations about offline versus online behavior.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill directs the agent to spawn four persistent subagent sessions and coordinate them over time, which materially expands its operational capability beyond simple stock analysis into autonomous orchestration. Even if intended for performance, persistent multi-agent control increases the risk of runaway actions, harder-to-audit behavior, and broader misuse if the workflow is repurposed or manipulated.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The entire skill is written as a Chinese-only workflow and prescribes a fixed Chinese response format, but it does not offer any language or locale choice to the user. Under the language/locale policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale restriction is clearly documented and justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The document title, role definition, and all task instructions are written exclusively in Chinese, and the generated report template also mandates Chinese output. Under the policy, forcing a specific language without user opt-in is a natural-language locale violation unless the constraint is explicitly justified, which it is not here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The entire skill is written as mandatory Chinese instructions and output templates, including fixed Chinese headings and evaluation text, but it does not provide any user opt-in or alternative language option. Under the policy, forcing a specific language without user choice is a natural-language policy violation unless the locale restriction is explicitly justified.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest describes the skill as an 'A股基本面深度分析工具', which implies company fundamentals analysis. This prompt is explicitly a '资金与市场情绪分析任务' and focuses on capital flows, dragon-tiger list activity, margin sentiment, and market attitude scoring, which are trading/sentiment dimensions rather than core fundamentals.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The prompt explicitly instructs the agent to make portfolio-style investment judgments such as '重仓', '轻仓', and '排除' without any risk disclosure, suitability caveat, or reminder that the output is informational rather than personalized financial advice. In the context of a stock-analysis skill, this increases the chance that users treat the model output as actionable investment advice, which can lead to financial harm or regulatory/compliance issues.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The entire guide, task labels, status messages, and report template are written as prescriptive Chinese-language content, including the fixed user reply at L094 and the required report structure at L102-L178. There is no indication that the user can choose another language or that the locale restriction is required for a region-specific compliance purpose, which creates a language/locale policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

This markdown skill mandates a single language for all instructions and outputs, which is a natural-language locale policy concern under the language-choice rule. There is no indication that the user can choose another language or that the Chinese-only requirement is justified as a region-specific constraint.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill instructs subagents to write scraped data to a specific local filesystem path without any user-facing disclosure or consent step. This can cause unexpected local state changes, consume disk space, and overwrite or create files in the operator's workspace, which is risky even if the intended purpose is benign financial analysis.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly creates a data directory and a large set of report artifacts, but does not notify the user that running it will modify local storage. In a multi-agent workflow, this increases the chance of unnoticed file proliferation and unintended persistence of scraped financial data in the local environment.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

该技能文档的标题、步骤说明和子Agent消息模板均仅以中文给出,没有提供用户语言选择、双语支持或限定其仅适用于中文工作流的说明。根据规则,强制单一语言而无用户选择可能构成语言/locale 策略问题。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

All user-facing instructions in this file are presented only in Chinese, and there is no indication that the skill supports other languages or that Chinese-only operation is an intentional, justified regional constraint. That creates a natural-language locale restriction without user opt-in.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

This markdown file instructs the agent to produce six output files and specifies they must exist in data/{代码}/, which implies filesystem writes. The documentation does not warn users that running the workflow will create or potentially overwrite files in that directory, so the behavior could affect user data without explicit disclosure.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The prompt explicitly instructs the agent to write output to a fixed local filesystem path. In an agent environment, this creates an unsafe side effect boundary because the model is being directed to modify local files without an explicit user confirmation, sandbox guarantee, or path restriction mechanism. The finance-analysis context makes this somewhat expected operational behavior, but it is still a real security concern because prompt-controlled file writes can be abused if variables or upstream inputs are manipulated.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The file is written as a prescriptive prompt in Chinese and includes the Traditional Chinese sentence 這個行業的天花板在哪裡?這家公司的壁壘能持續多久? within otherwise Simplified Chinese content. This imposes a language/locale style without any user opt-in or documented reason, which can conflict with language choice policy.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

This markdown skill explicitly directs the agent to write results to a fixed path under the workspace. Because file modification can affect user data, the skill description should disclose that the target file will be created or overwritten if it already exists.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The natural-language instructions, headings, and report template are all in Chinese, which effectively fixes the interaction/output language without any user opt-in. Under the language/locale policy rule, forcing a specific language is a policy concern unless the file documents a justified locale constraint or offers a choice.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

This markdown file explicitly defines an output file path, implying the skill will write generated analysis to the user's workspace. The description does not include any warning or disclosure that running the skill will create or overwrite a local file, which is a user-impacting filesystem action under the markdown warning criteria.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The prompt instructs the agent to write an output file in the workspace without any user-facing disclosure or confirmation. In an agent environment, silent file creation/modification can violate user expectations and reduce auditability, especially when triggered implicitly from a stock-analysis request.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The task adds '产业链关联分析' and '重大事项追踪', including listing related companies and judging stock-price impact from announcements. While adjacent to equity research, these sections broaden the behavior from pure fundamental analysis into event-driven and relationship mapping work that is not stated in the manifest.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

All instructions, labels, and expected outputs are specified exclusively in Chinese, and the skill does not indicate that the user may select another language. Under the policy, forcing a specific language without user opt-in is a natural-language policy concern unless the locale restriction is explicitly documented and justified.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly instructs the agent to write a file to a fixed local path, which causes filesystem side effects without any user-facing confirmation or clear boundary checks. Even though this appears to be part of the normal workflow for this analysis skill, silent file writes can overwrite existing data or be abused if upstream inputs such as 股票代码 are not tightly validated.

Content

No source excerpt is available for this finding.