Back to skill

Security audit

Yezhou Deploy

Security checks for vulnerabilities and agentic risk

Overview

This skill mostly does what it claims, but an undocumented endpoint override could send a saved Yezhou token and project contents to a non-Yezhou server if the environment is manipulated.

Install only if you trust this publisher and the environment where agents will run it. Before use, check that YEZHOU_BASE_URL is unset or set only to the intended Yezhou service, because this skill stores a Yezhou credential locally and uploads the selected web project to the configured server.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/publish.mjs:31
Finding

Arbitrary API Endpoint Override Can Expose Credentials and Project Data

Content
View full analysis
({})); return { response, body }; } ``` The saved bearer credential is sent through this helper when listing sites: ```js async function listSites(accessToken){ return api("/api/agent/sites",{headers:{Authorization:`Bearer ${accessToken}`}}); } ``` It is also sent with selected project content when publishing or updating a site: ```js async function publish(accessToken, payload, siteId) { const project=inputKind!=="html",basePath=project?"/api/agent/projects":"/api/agent/sites"; return api(siteId ? `${basePath}/${encodeURIComponent(siteId)}` : basePath, { method:siteId ? "PUT" : "POST", headers:{Authorization:`Bearer ${accessToken}`,...(project?{}:{"Content-Type":"application/json; charset=utf-8"})}, body:project?payload:JSON.stringify(payload), }); } ``` ### Technical Analysis `YEZHOU_BASE_URL` can specify an arbitrary origin, including an attacker-controlled HTTPS server, a plaintext HTTP server, or a local-network endpoint. No origin allowlist, HTTPS requirement, or separation of credentials by origin is implemented. After reading the bearer credential from the user's configuration directory, the script attaches it to requests sent to the configured origin. Publ ...[truncated 2656 chars]
Remediation
View remediation
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
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (10)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 13)May include surrounding context.

md
node scripts/publish.mjs <html-file|directory|zip-file> [title]

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 19)May include surrounding context.

md
node scripts/publish.mjs <html-file|directory|zip-file> [title]

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 20)May include surrounding context.

md
node scripts/publish.mjs <html-file|directory|zip-file> [title]

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
90% confidence
Finding

The skill invokes a bundled Node.js script that performs shell execution, networked publishing, and likely accesses local environment/configuration, but the manifest declares no explicit tool scope or permissions. This creates an under-specified trust boundary: an agent may run code with broader capabilities than a reviewer or user expects, increasing the risk of unintended file access, outbound uploads, or command execution.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 25)May include surrounding context.

md
Use `--site` only when the user identifies the existing work to update. It accepts either the 12-character site ID or its full `/p/<id>` public URL. A successful update writes the recovered binding into the local `.yezhou.json`, so later publishes on that computer need no extra option. Never guess which remote work should be overwritten.

On first use, the script starts 页舟 browser authorization. Tell the user to approve the matching one-time code in the browser, then continue waiting for the script. Do not ask the user to create, copy, paste, or expose a token. The credential is stored in the user's configuration directory, never in the project.

On the first publish, the script creates one `.yezhou.json` manifest in the input's project directory. The manifest keeps separate bindings keyed by the directory or HTML/ZIP filename, so sibling inputs do not share a site ID. On later runs with the same input, it updates the recorded site ID, so the public URL stays unchanged. The first use of a legacy single-binding manifest migrates that binding to the current input automatically. Return the URL printed by the script.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 31)May include surrounding context.

md
Treat `.yezhou.json` as project metadata. The publisher never creates, edits, or overwrites `.gitignore`, Git configuration, or other project files. It writes only its `.yezhou.json` binding manifest. If the user wants that manifest ignored or tracked, leave the choice and Git change to them. The state file contains no secret, but publishing it can expose internal project identifiers. Do not replace or delete an existing binding unless the user explicitly asks to publish as a new site. If browser authorization cannot be opened automatically, show the verification URL and one-time code printed by the script.

The publisher requires Node.js 18 or newer and works on Windows, macOS, and Linux. A directory must contain `index.html` at its root. A ZIP may contain `index.html` at the archive root or inside one common top-level wrapper folder, which the server removes automatically. Directories are uploaded automatically; do not ask the user to create a ZIP first. Hidden files and `node_modules` are excluded. SVG is intentionally rejected instead of silently omitted because serving active SVG on the content origin would create a script-execution risk; convert it to PNG/WebP or a safe HTML/CSS icon. Markdown is not supported.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This code packages local files or HTML content and sends them to remote API endpoints using fetch, including all non-hidden files in a directory and optional titles. While publishing is the script's purpose, the file itself provides no explicit comment, docstring, or user-facing disclosure that local content will be uploaded to a remote host, which is the kind of network data transmission this rule asks to flag when no warning is present in the code file.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

The skill title and surrounding instructions are presented in Chinese and centered on a specific platform without stating whether users may interact in another language or locale. Under the policy rule, a skill should not implicitly force a language or locale unless that constraint is documented and justified or the user is given a choice.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The short description and default prompt are written only in Chinese, which can constitute a language/locale policy violation when no user opt-in or documented locale restriction is provided. The file does not indicate that the skill is region-specific or that users may choose another language.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The script uses Chinese-only defaults and labels such as '未知设备' and '页舟 Deploy', and elsewhere emits multiple Chinese user-facing messages. There is no indication that the user can opt into another language or locale, which matches the policy concern for forced language behavior.

Content

No source excerpt is available for this finding.

Static analysis

Detected: suspicious.dangerous_exec, suspicious.potential_exfiltration

Shell command execution detected (child_process).

Critical
Code
suspicious.dangerous_exec
Location
scripts/publish.mjs:60

Sensitive-looking file read is paired with a network send.

Warn
Code
suspicious.potential_exfiltration
Location
scripts/publish.mjs:96