Back to skill

Security audit

coze-workflow-runner

Security checks for vulnerabilities and agentic risk

Overview

This Coze workflow skill is mostly purpose-aligned, but it uses local service credentials to run arbitrary remote workflows and documents unvalidated downloads from workflow output.

Review before installing. Use only with a least-privilege Coze token, restrict allowed workflow IDs, avoid sending sensitive personal or business data unless users consent, and do not automatically download workflow-returned URLs without validating the host, redirects, file type, size, and destination filename.

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)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
scripts/run_workflow.py:18
Finding

Shared Service Credential Permits Unrestricted Workflow Invocation

Content
View full analysis
2 else "测试" print(f"调用工作流 {workflow_id} ...") result = run_workflow(workflow_id, input_text) ``` ### Technical Analysis The script reads a shared Coze service credential from a fixed local file and uses it to execute a workflow ID supplied directly through the command line. It does not verify that the requested workflow is one of the workflows documented or approved by the Skill. Consequently, the effective authorization boundary is determined entirely by the permissions of the service token rather than by the caller's intended permissions. Any caller able to invoke the script can attempt to execute other workflows ava ...[truncated 1640 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:91
Finding

Unvalidated Workflow URLs Are Downloaded with Redirect Following

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (12)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding

The skill references sensitive local files such as ~/workspace-prod/coze/coze-tokens.md and ~/workspace-prod/coze/workflows.md but does not declare any explicit tool scope or permission boundaries. In an agent setting, undeclared file-read behavior can cause the agent to access credentials or configuration outside user expectations, increasing the chance of secret exposure or overbroad filesystem access.

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 use of a service token and submission of inputs to the external Coze API without warning that user data will be transmitted to a third-party service or that credentials are being handled locally. This can lead to inadvertent disclosure of sensitive prompts, links, or business data, especially in environments where users assume the skill operates locally.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The documentation instructs users to invoke curl through subprocess on workflow-returned URLs, causing the local environment to fetch arbitrary external content and write it to disk. Because these URLs originate from workflow output, this creates a trust boundary violation: a compromised or misconfigured workflow could return attacker-controlled links that trigger unwanted network access, content download, or overwriting of local files.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill documents automatic retrieval of external URLs and saving them as local files without warning users that it will create files and download remote content. This reduces informed consent and can expose users to malicious files, excessive downloads, or storage abuse if returned links are unexpected or hostile.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The document instructs sending user email addresses and related personal data to an external Coze workflow API without any privacy notice, consent language, data handling constraints, or minimization guidance. This creates a real privacy and compliance risk because operators may transmit personally identifiable information to a third party without adequately informing users or restricting what is sent.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
84% confidence
Finding

The file documents forwarding user-supplied workflow parameters, including email content, to an external domain (api.coze.cn). External transmission is not inherently malicious, but in this context it is security-relevant because the example proxy blindly relays data to a third-party service without documenting validation, allowlisting, consent, or redaction controls.

Content

Scanner excerpt · references/workflows.md (reported line 26)May include surrounding context.

后端代理(Node.js)

javascript
const COZE_API_URL = 'https://api.coze.cn/v1/workflow/run';
const COZE_AUTH_TOKEN = 'Bearer 你的服务令牌';

app.post('/api/proxy/coze', async (req, res) => {

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The script reads a bearer token from a hard-coded absolute path in the user's home directory and uses it automatically for API access. This creates unauthorized credential-access capability beyond merely accepting a token as input, ties execution to a specific local secret file, and risks misuse or unintended disclosure if the skill is run in another user's environment or repurposed by an agent without explicit consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The script silently accesses a local credential file and extracts a bearer token without any user-facing notice or consent prompt. In an agent skill context, hidden access to sensitive local credentials is especially risky because users may believe the skill only runs a workflow, while it actually harvests and reuses an existing secret from disk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

The natural-language instructions and examples are entirely in Chinese, which can amount to forcing a specific language without user opt-in when no locale constraint or alternative is provided. SQP-3 covers language-policy issues of this type across all file types.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

All visible instructions, labels, and examples are presented only in Chinese, with no indication that other languages are supported or that Chinese is a justified regional requirement. Under the language/locale policy rule, forcing a single language without opt-in can be a policy concern.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The manifest says the skill runs Coze workflows for broad automation tasks including image generation and data processing, but this file documents only a specific workflow for receiving an email address and triggering downstream business logic. This suggests the documented behavior in the skill materials is materially narrower than the declared capability surface.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The natural-language docstring, usage guidance, and runtime messages are all Chinese, which effectively forces a specific language for users. The file does not offer a language/locale choice or explain that the skill is intentionally limited to a Chinese-speaking context.

Content

No source excerpt is available for this finding.