Back to skill

Security audit

市场调研分析助手

Security checks for vulnerabilities and agentic risk

Overview

This market research skill is mostly purpose-aligned, but its fallback crawler can make unscoped web requests to URLs from search results and feed untrusted page text back into the agent.

Review before installing. The skill is not clearly malicious, but run its crawler only in a sandboxed environment with restricted outbound network access, avoid using it where internal services or metadata endpoints are reachable, and treat all crawled text as untrusted source material rather than instructions. Prefer the RSS/web_fetch path and only use -o output when you intentionally want a local JSON file.

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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/research_crawler.py:161
Finding

Server-Side Request Forgery Through Unvalidated Search Result URLs

Content
View full analysis
Remediation
View remediation

T01 · Skill Instruction Hijacking

Warning
Location
scripts/research_crawler.py:179
Finding

Indirect Prompt Injection Through Untrusted Crawled Web Content

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (11)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The stated purpose is market analysis, but the instructions also direct active web crawling, search-engine querying, content extraction, and local JSON output. This mismatch is dangerous because it hides materially riskier behavior behind an innocuous description, reducing informed consent and making misuse or over-privileged execution more likely.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding

The skill instructs the agent to perform network access and potentially write files via the crawler output, but it declares no explicit tool scope or permissions. This creates an authorization gap where a seemingly simple analysis skill can invoke higher-risk capabilities without transparent user or platform constraints.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The manifest description says the skill triggers on requests for '市场调研、行业分析、竞争分析、商业可行性分析等', where the '等' and broad analysis terms make activation boundaries unclear. This lacks specific scope limits or negative examples, increasing the chance of unintended invocation for generic business-analysis requests.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill includes instructions to run a local Python crawler script as a fallback, expanding capability from text analysis into arbitrary code execution. Running local scripts increases attack surface substantially because the script may access the filesystem, network, or other local resources beyond what users expect from a research assistant.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The instructions describe network crawling and saving results to local files without explicit user warning or opt-in. This is risky because it can create unexpected outbound requests and local artifacts, affecting privacy, auditability, and user trust.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The HTTP header sets Accept-Language to "zh-CN,zh;q=0.9,en;q=0.8", which forces a Chinese locale preference for all requests. The file also presents all user-facing help text in Chinese, but does not state that the tool is region-specific or allow users to opt into another language/locale.

Content

No source excerpt is available for this finding.

Tainted flow: 'href' from requests.get (line 100, network input) → requests.get (network output)

Medium
Category
Data Flow
Confidence
90% confidence
Finding

The code follows a URL extracted from untrusted Baidu search result HTML with requests.get(..., allow_redirects=True) and no hostname or scheme validation. This creates an SSRF-style outbound request primitive that can be abused to make the agent contact arbitrary external endpoints and potentially internal services if attacker-controlled content appears in results or redirect chains.

Content

Scanner excerpt · scripts/research_crawler.py (reported line 99)May include surrounding context.

python
# 百度搜索结果需要解析真实URL
                if 'baidu.com' in href and 'http' not in href:
                    try:
                        resp = requests.get(href, headers=HEADERS, timeout=5, allow_redirects=True)
                        href = resp.url
                    except:
                        pass

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The manifest describes a market-research assistant that generates structured analysis from a user-provided topic, but this script directly fetches arbitrary URLs and extracts full page content. While web retrieval is relevant to research, implementing a standalone crawler that downloads and parses arbitrary pages is a materially broader behavior than the manifest's analysis-focused description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

When search engines fail, the code fetches homepage content from hardcoded news sources and returns it as research input, even though those pages may not match the requested industry or product topic. This diverges from the manifest's promise to analyze the user-provided research subject, because the fallback can supply generic site content instead of topic-specific market research.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instructions specify '默认中文输出' and require Chinese punctuation/formatting, which imposes a language preference without indicating that the user may choose another language. Per the policy, forcing a specific language without opt-in is a natural-language policy issue unless clearly justified or optional.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest describes generating a market analysis report, but the code also supports saving raw search and crawl output to an arbitrary local JSON file. Local file output is not obviously implied by the skill description and represents extra behavior beyond simply producing analysis for the user.

Content

No source excerpt is available for this finding.