Back to skill

Security audit

ai-workflow-engine

Security checks for vulnerabilities and agentic risk

Overview

This workflow skill has a coherent automation purpose, but it grants broad network, file, database, email, and plugin-loading authority without enough scoping or safety controls.

Install only if you intend to run a broad automation framework and can review generated workflows before execution. Avoid giving it sensitive records, credentials, private network URLs, or email/database write access unless you have destination allowlists, dependency pinning, isolated execution, and explicit approval gates in place.

Vulnerability Patterns
  • Tool Hijacking and SpoofingModifies or replaces tools so legitimate-looking calls execute attacker logic
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (3)

T09 · Insecure Skill Coding Practices

Error
Location
workflow_enhanced.py:115
Finding

Unrestricted Outbound Requests Enable SSRF and Attacker-Directed Data Transmission

Content
View full analysis
str: """获取URL内容""" print(f"🌐 请求: {url}") try: import requests resp = requests.get(url, timeout=30) resp.encoding = encoding or resp.apparent_encoding return resp.text except Exception as e: print(f"请求失败: {e}") return "" ``` ### Technical Analysis Both functions make network requests to caller-supplied URLs without validating the destination. There is no restriction on URL schemes, destination hosts, resolved IP addresses, ports, or redirects. In particular, the code does not reject loopback, private, reserved, or link-local addresses. `WebhookTrigger.call_webhook()` presents the greater exposure because it also permits caller-controlled HTTP methods, headers, and JSON bodies. It has no request timeout or response-size restriction. Consequently, data passed to this function can be transmitted to any destination reachable from the host. Network access is consistent with the Skill's advertised scraping, webhook, and AI integration functionality. However, unrestricted access to arbitrary internal and external destinations exceeds the minimum privilege necessary for those features. The implementation should limit requests to explicitly approved destinations. The code does not independently enumerate credentials or automatically transmit local files. Exploitation requires untrusted input to influence the UR ...[truncated 1482 chars]
Remediation
View remediation

T07 · Tool Hijacking and Spoofing

Error
Location
ai_workflow.py:26
Finding

Hard-Coded Search-Path Modification Allows External Tool Hijacking

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
requirements.txt:1
Finding

Unpinned Third-Party Dependencies Create a Mutable Supply-Chain Risk

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • 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 (31)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The headline and usage framing present the skill in Chinese only, with no indication that users may choose another language or that the locale restriction is intentional for a region-specific tool. The policy requires flagging language or locale constraints when they are imposed without user opt-in or justification.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This markdown file describes workflows that write Excel files, insert data into databases, and fetch or scrape remote URLs, but it does not include any warning about modifying user data, persisting records, or making network requests. For markdown files, omitted disclosure of behaviors affecting user data, privacy, or system integrity should be flagged.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill encourages automatic workflow execution that includes web scraping, database writes, report generation, and email sending, but it does not clearly warn users that invoking the generated workflow can cause immediate external side effects. In an agent environment, this can lead to unintended network access, outbound communications, and persistent changes without informed user consent or a confirmation step.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The Agent examples grant tools capable of writing files and sending email but present them as normal capabilities without disclosing the risks of autonomous tool use. This is dangerous because users may enable agents that perform external actions or data exfiltration paths without understanding the permission scope or adding safeguards.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The code-generation section states that generated code will automatically include database operations, file export, and email sending, but it omits a warning that such generated code may execute impactful real-world actions. This raises the risk that users will run generated workflows that write to databases, create files, or send emails without review, validation, or approval, amplifying harm because the actions are synthesized automatically from natural-language prompts.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This Python file contains its primary module description in Chinese and presents itself as a general-purpose workflow engine, but it does not indicate that Chinese is optional or that the skill is region-specific. That creates a language/locale policy issue because the skill implicitly fixes the interaction language without user opt-in.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The docstrings and method names in ExcelWrapper.write and ExcelSteps.write_excel indicate that caller-supplied data will be written to an Excel file. In reality, the code instantiates a parser, creates a sheet, saves, and closes without ever using the data argument, which directly contradicts the documented intent of the function.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The module-level docstring describes an AI workflow engine focused on workflow orchestration, agent collaboration, RAG, data processing, Excel, databases, and charts. The separate NetworkSteps class adds general-purpose HTTP retrieval and web scraping, which is not mentioned in that stated scope and materially expands the skill's capabilities beyond the documented purpose.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This Python file contains prominent natural-language descriptions entirely in Chinese, including the module docstring and many user-facing strings, with no indication that language selection is optional. Under the language/locale policy, forcing a specific language without user opt-in is a policy concern 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
95% confidence
Finding

The prompt template in NLCodeGenerator is written entirely in Chinese and directly instructs how code should be generated, implying a fixed language context for interaction. Because no alternative language or opt-in path is offered, this is a natural-language locale policy violation.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The prompt template in L630-L634 states generated code should include error handling and logging, but the implementation returned by _template_generate only emits simple print-based stub functions and no try/except or logging setup. This is an active contradiction between the documented intent of the generator and the code it actually produces.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The file's title, docstrings, comments, and user-visible text are predominantly in Chinese, with no indication that the user can choose language or that the skill is intentionally region-specific. This creates a language/locale policy concern because the skill effectively enforces a specific language without opt-in.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

The presence of a hardcoded external API endpoint confirms intentional egress to a third-party service. While this may be expected functionality, it is still a true security concern in a workflow skill because data can leave the trusted environment and there is no surrounding disclosure, approval, or restriction logic.

Content

Scanner excerpt · workflow_enhanced.py (reported line 53)May include surrounding context.

python
def _deepseek_chat(self, prompt, model="deepseek-chat", **kwargs):
        try:
            import requests
            resp = requests.post("https://api.deepseek.com/v1/chat/completions", 
                headers={"Authorization": f"Bearer {self.api_key}"},
                json={"model": model, "messages": [{"role": "user", "content": prompt}]})
            return resp.json()

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

The presence of a hardcoded external API endpoint confirms intentional egress to a third-party service. While this may be expected functionality, it is still a true security concern in a workflow skill because data can leave the trusted environment and there is no surrounding disclosure, approval, or restriction logic.

Content

Scanner excerpt · workflow_enhanced.py (reported line 53)May include surrounding context.

python
def _deepseek_chat(self, prompt, model="deepseek-chat", **kwargs):
        try:
            import requests
            resp = requests.post("https://api.deepseek.com/v1/chat/completions", 
                headers={"Authorization": f"Bearer {self.api_key}"},
                json={"model": model, "messages": [{"role": "user", "content": prompt}]})
            return resp.json()

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

The presence of a hardcoded external API endpoint confirms intentional egress to a third-party service. While this may be expected functionality, it is still a true security concern in a workflow skill because data can leave the trusted environment and there is no surrounding disclosure, approval, or restriction logic.

Content

Scanner excerpt · workflow_enhanced.py (reported line 53)May include surrounding context.

python
def _deepseek_chat(self, prompt, model="deepseek-chat", **kwargs):
        try:
            import requests
            resp = requests.post("https://api.deepseek.com/v1/chat/completions", 
                headers={"Authorization": f"Bearer {self.api_key}"},
                json={"model": model, "messages": [{"role": "user", "content": prompt}]})
            return resp.json()

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The webhook helper permits arbitrary outbound HTTP requests to attacker-controlled URLs with arbitrary payloads and headers, with no validation, allowlist, authentication policy, or user disclosure. In a skill/workflow context, this can exfiltrate sensitive workflow data or turn the agent into an SSRF-capable HTTP client if untrusted inputs can influence the URL or body.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill's natural-language instructions and examples are written in Chinese, which effectively forces a specific language for users. There is no indication that users can opt into another language or that the skill is intentionally restricted to a Chinese-language audience for a documented reason.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
98% confidence
Finding

The dependency list leaves pandas unpinned, which makes builds non-reproducible and allows future installs to silently pull different releases, including vulnerable or breaking ones. While this file alone does not prove exploitation, it weakens supply-chain control and makes risk assessment difficult.

Content

Scanner excerpt · requirements.txt (reported line 1)May include surrounding context.

text
pandas
requests
openai
anthropic

Unverifiable Dependency: pandas has 1 known advisory(ies) (CVE-2020-13091 (** DISPUTED ** pandas through 1.0.3 can unserialize and execute commands from an)), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
89% confidence
Finding

Pandas has at least one known advisory, but because the version is not pinned there is no way to determine whether the installed build is affected. The immediate issue is not confirmed exploitation of a specific CVE, but an inability to verify safety of the dependency set.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
98% confidence
Finding

The requests package is unpinned, so installations may resolve to different versions over time, including versions with known security defects or incompatible behavior. This creates avoidable supply-chain uncertainty for any code that performs network access.

Content

Scanner excerpt · requirements.txt (reported line 2)May include surrounding context.

text
pandas
requests
openai
anthropic
chromadb

Unverifiable Dependency: requests has 16 known advisory(ies) (CVE-2014-1830 (Exposure of Sensitive Information to an Unauthorized Actor in Requests); CVE-2024-47081 (Requests vulnerable to .netrc credentials leak via malicious URLs); CVE-2024-35195 (Requests `Session` object does not verify requests after making first request wi) +13 more), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
95% confidence
Finding

Requests has multiple known advisories and the missing version pin means the environment could resolve to a vulnerable release without visibility. Because requests is a core HTTP client, affected versions can directly impact outbound authentication, TLS behavior, and credential handling.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
98% confidence
Finding

The openai dependency is not version-pinned, which means future installs may unexpectedly pull newer releases with changed behavior or newly introduced flaws. Unpinned SDKs are especially risky in agent skills because they often mediate remote model access and tool behavior.

Content

Scanner excerpt · requirements.txt (reported line 3)May include surrounding context.

text
pandas
requests
openai
anthropic
chromadb
pypdf

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
98% confidence
Finding

The anthropic package is unpinned, so dependency resolution can drift to arbitrary future releases. This increases supply-chain risk and prevents confident verification against known advisories.

Content

Scanner excerpt · requirements.txt (reported line 4)May include surrounding context.

text
pandas
requests
openai
anthropic
chromadb
pypdf
beautifulsoup4

Unverifiable Dependency: anthropic has 4 known advisory(ies) (CVE-2026-34450 (Claude SDK for Python has Insecure Default File Permissions in Local Filesystem ); CVE-2026-34452 (Claude SDK for Python: Memory Tool Path Validation Race Condition Allows Sandbox); CVE-2026-34450 (The Claude SDK for Python provides access to the Claude API from Python applicat) +1 more), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
94% confidence
Finding

Anthropic has known advisories and the unpinned manifest prevents verification that a safe version will be installed. In an agent skill context, SDK issues can affect local file handling, sandbox boundaries, or tool interactions, increasing operational risk.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
99% confidence
Finding

The chromadb package is unpinned, which is more concerning because databases/vector stores can expose data access and code-execution attack surface depending on version and deployment mode. Without pinning, the environment may install a vulnerable release and make incident triage much harder.

Content

Scanner excerpt · requirements.txt (reported line 5)May include surrounding context.

text
requests
openai
anthropic
chromadb
pypdf
beautifulsoup4
lxml