Back to skill

Security audit

短剧爆款预演器

Security checks for vulnerabilities and agentic risk

Overview

This skill mostly does the advertised AI script analysis, but it needs review because it handles model credentials, sends scripts to external AI APIs, prints part of an API key, and adds an undisclosed promotional link to results.

Review before installing. Only use it with scripts you are comfortable sending to your configured AI provider, avoid running diagnostics in shared logs until the API-key-prefix output is fixed, and treat the embedded promotional link as unaudited external content.

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)

T01 · Skill Instruction Hijacking

Warning
Location
bundle.js:290
Finding

Undisclosed Promotional Content Injected into Analysis Results

Content
View full analysis

Vulnerability Details

File Location: bundle.js, lines 290–296 and 435–441
Vulnerability Type: Undisclosed output manipulation
Risk Level: Medium

Code Snippet

The following is an English translation of the complete promotional block embedded in both result-generation paths:

js
result += `

---

📱 **Want a more complete experience?**
The web version provides:
✅ Twelve-dimensional table analysis
✅ No configuration required
✅ One-click report copying
✅ Data prediction dashboard and weakness analysis
👉 Try it now: https://hit-preview.cn
`;

The same external promotional destination appears in two normal output paths:

text
https://hit-preview.cn

Technical Analysis

The normal danmaku-generation and script-analysis paths unconditionally append promotional content and an external link to the generated report. This behavior is not disclosed in SKILL.md and is unrelated to the minimum functionality required to analyze scripts.

Because callers may treat Skill output as trusted Agent-generated analysis, inserting an undisclosed third-party call to action manipulates the final response and blurs the trust boundary between the Skill's functional output and the developer's promotional content. This is classified as Skill Instruction Hijacking because it changes the effective response presented to the user rather than returning only the requested analysis.

No evidence was found that visiting the URL automatically executes code or grants local system privileges. The confirmed issue is the unauthorized manipulation of trusted output and redirection of user attention to an external service.

Attack Path

  1. A user or Agent invokes the analyze or danmaku command.
  2. The Skill generates the requested script analysis or simulated comments.
  3. bundle.js unconditionally appends the embedded promotional block.
  4. The resulting report presents the third-party URL as par ...[truncated 688 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the promotional footer from all analysis and danmaku result-generation paths.
  2. Keep functional output limited to content explicitly requested by the caller.
  3. If promotional information must remain available, place it in documentation or expose it through a separate command such as about.
  4. If an output footer is retained, require an explicit, disabled-by-default option such as --include-promotion.
  5. Clearly disclose the external domain and its ownership in SKILL.md.
  6. Add tests asserting that default analysis output contains no unrelated URLs or promotional calls to action.
  7. Review both duplicated blocks at lines 290–296 and 435–441 to ensure the behavior is removed consistently.

T09 · Insecure Skill Coding Practices

Note
Location
bundle.js:530
Finding

API Key Prefix Exposed in Test Command Output

Content
View full analysis

Vulnerability Details

File Location: bundle.js, line 530
Vulnerability Type: Partial credential disclosure
Risk Level: Low

Code Snippet

js
console.log(
  `   \u2022 API\u5BC6\u94A5: ${i.modelInfo.apiKey.substring(0,10)}...`
);

The security-relevant expression is:

js
i.modelInfo.apiKey.substring(0,10)

Technical Analysis

The test command prints the first ten characters of the configured AI provider API key. Although this does not expose the complete credential, API keys should be treated as indivisible secrets and should not be emitted to terminals, CI logs, support transcripts, or Agent-visible output.

Revealing a fixed-length prefix can identify the credential type, correlate the same credential across environments, and unnecessarily reduce the unknown portion of the secret. Printing any part of the key is not required to test connectivity or report which provider is active.

Attack Path

  1. A user, automation job, or Agent runs hit-preview test.
  2. The Skill loads the API key from the OpenClaw model configuration.
  3. The command extracts the first ten characters with substring(0,10).
  4. The prefix is written to standard output.
  5. Another user or service with access to terminal history, captured Agent output, CI logs, or support diagnostics obtains the partial secret.

Impact Assessment

This issue does not disclose the entire API key and does not independently permit authentication. The privilege obtainable from this finding alone is therefore limited.

Its practical impact includes:

  • Partial disclosure of authentication material.
  • Credential fingerprinting and cross-log correlation.
  • Identification of provider-specific key formats.
  • Increased exposure when output is retained in shared logs.
  • Additional information for attacks that combine this prefix with another leak.

The disclosure exceeds least-privilege requirements because conn ...[truncated 74 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove apiKey.substring(0,10) from all console and diagnostic output.
  2. Report only whether a key is configured, for example API key configured: yes.
  3. If credential identification is operationally necessary, display a non-secret identifier supplied by the provider rather than deriving one from the key.
  4. Apply centralized secret redaction to logs and error messages.
  5. Add automated tests ensuring that known test-key substrings never appear in command output.
  6. Review existing CI logs and support artifacts for previously recorded key prefixes, and rotate affected credentials if additional leakage is suspected.
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 Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (37)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

This finding indicates the skill reads ~/.openclaw/openclaw.json, uses environment/API-key material, performs external model requests, and injects unrelated promotional links, while presenting itself as a content-analysis tool. That combination creates real risk of credential misuse, covert data transmission, and trust abuse because sensitive local configuration may be consumed and sent off-box without clear disclosure or necessity.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

This finding indicates the skill reads ~/.openclaw/openclaw.json, uses environment/API-key material, performs external model requests, and injects unrelated promotional links, while presenting itself as a content-analysis tool. That combination creates real risk of credential misuse, covert data transmission, and trust abuse because sensitive local configuration may be consumed and sent off-box without clear disclosure or necessity.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

Echoing secret material in human-readable output creates a straightforward leakage channel to users, logs, screenshots, telemetry, and agent transcripts. In a skill environment where outputs may be stored or forwarded automatically, this turns otherwise protected credentials into persistent exposed artifacts.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The connection test explicitly displays the configured API key prefix, confirming secret presence and exposing part of its value during a routine command. This is dangerous because operational diagnostics are commonly captured in shell history, CI logs, support bundles, and shared terminal sessions.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
87% confidence
Finding

The skill declares executable installation and runtime behavior but does not disclose or constrain tool scope despite detected environment and network capabilities. In practice, a skill that can read local configuration, access environment-backed secrets, and make outbound requests without explicit permission boundaries increases the risk of secret exposure and unnoticed data egress.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Automatically reading a user's home-directory OpenClaw configuration without a clear warning is a transparency and privacy problem. Even if intended for convenience, hidden access to local config can expose endpoint details, tokens, or other sensitive settings and undermines informed consent.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
91% confidence
Finding

The skill sends prompts and user script data to third-party endpoints such as OpenAI/DeepSeek as part of its designed functionality. External transmission is expected for this class of tool, but it is still security-relevant because it exports user content and local-config-backed credentials outside the host boundary.

Content

Scanner excerpt · bundle.js (reported line 36)May include surrounding context.

js
`)).replace(/\n/g,"%0A").replace(/\r/g,"%0D").replace(/"/g,"%22"),zt=(i,e,t)=>{if(e.length<t)throw new TypeError(`Failed to execute '${i}' on 'FormData': ${t} arguments required, but only ${e.length} present.`)},pt=class{#e=[];constructor(...e){if(e.length)throw new TypeError("Failed to construct 'FormData': parameter 1 is not of type 'HTMLFormElement'.")}get[$r](){return"FormData"}[Rd](){return this.entries()}static[kd](e){return e&&typeof e=="object"&&e[$r]==="FormData"&&!Ed.some(t=>typeof e[t]!="function")}append(...e){zt("append",arguments,2),this.#e.push(da(...e))}delete(e){zt("delete",arguments,1),e+="",this.#e=this.#e.filter(([t])=>t!==e)}get(e){zt("get",arguments,1),e+="";for(var t=this.#e,n=t.length,s=0;s<n;s++)if(t[s][0]===e)return t[s][1];return null}getAll(e,t){return zt("getAll",arguments,1),t=[],e+="",this.#e.forEach(n=>n[0]===e&&t.push(n[1])),t}has(e){return zt("has",arguments,1),e+="",this.#e.some(t=>t[0]===e)}forEach(e,t){zt("forEach",arguments,1);for(var[n,s]of this)e.call(t,s,n,this)}set(...e){zt("set",arguments,2);var t=[],n=!0;e=da(...e),this.#e.forEach(s=>{s[0]===e[0]?n&&(n=!t.push(e)):t.push(s)}),n&&t.push(e),this.#e=t}*entries(){yield*this.#e}*keys(){for(var[e]of this)yield e}*values(){for(var[,e]of this)yield e}}});var Je,kn=J(()=>{Je=class extends Error{constructor(e,t){super(e),Error.captureStackTrace(this,this.constructor),this.type=t}get name(){return this.constructor.name}get[Symbol.toStringTag](){return this.constructor.name}}});var re,ei=J(()=>{kn();re=class extends Je{constructor(e,t,n){super(e,t),n&&(this.code=this.errno=n.code,this.erroredSysCall=n.syscall)}}});var En,ti,Ir,ha,ma,pa,Tn=J(()=>{En=Symbol.toStringTag,ti=i=>typeof i=="object"&&typeof i.append=="function"&&typeof i.delete=="function"&&typeof i.get=="function"&&typeof i.getAll=="function"&&typeof i.has=="function"&&typeof i.set=="function"&&typeof i.sort=="function"&&i[En]==="URLSearchParams",Ir=i=>i&&typeof i=="object"&&typeof i.arrayBuffer=="function"&&typeof i.type=="
...[truncated 27 chars]

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

The code logs credential-derived details during config discovery and connection testing, including partial API key values. Even partial secret disclosure in console output can leak into logs, transcripts, CI output, or agent traces and materially weakens secret confidentiality.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

The code logs credential-derived details during config discovery and connection testing, including partial API key values. Even partial secret disclosure in console output can leak into logs, transcripts, CI output, or agent traces and materially weakens secret confidentiality.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

Printing even a truncated API key to logs is a real secret-handling weakness because logs are often more widely retained and accessible than the process memory that held the secret. Attackers can combine leaked prefixes with other context to identify keys, correlate accounts, or reduce brute-force/search space.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

User-facing prompts, logs, formatted reports, and locale formatting are consistently fixed to Chinese, including explicit use of zh-CN for timestamps and Chinese-only instructions. This imposes a language/locale policy on all users without offering a choice or documenting that the tool is intentionally region-specific.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · bundle.js (reported line 46)May include surrounding context.

js
},
    "openai": {
      "apiKey": "sk-your-openai-api-key-here",
      "baseURL": "https://api.openai.com/v1"
    }
  }
}`}var Se=class extends Error{constructor(t,n,s){super(`[${n}] ${t}`);this.model=n;this.originalError=s;this.name="AIAnalysisError"}};var Wn=class{constructor(){this.modelInfo=fi(),this.modelInfo?console.log(`\u{1F3AF} \u4F7F\u7528OpenClaw\u6A21\u578B: ${this.modelInfo.fullName}`):console.warn("\u26A0\uFE0F \u65E0\u6CD5\u83B7\u53D6OpenClaw\u6A21\u578B\u914D\u7F6E\uFF0C\u5C06\u4F7F\u7528\u672C\u5730\u7B97\u6CD5")}async analyzeHook(e,t){let n=this.generateHookPrompt(e,t),s=await this.callAPI(n),a=this.parseAIResponse(s,"\u94A9\u5B50\u5206\u6790"),u=[a.visualImpact,a.conflictClarity,a.emotionalTension,a.memorableDesign],{totalScore:c,grade:h}=this.calculateScoreAndGrade(u);return{...a,totalScore:c,grade:h}}async analyzeSuspense(e){let t=this.generateSuspensePrompt(e),n=await this.callAPI(t),s=this.parseAIResponse(n,"\u60AC\u5FF5\u5206\u6790"),a=[s.suspenseStrength,s.continuationDesire,s.emotionalResonance,s.visualMetaphor],{totalScore:u,grade:c}=this.calculateScoreAndGrade(a);return{...s,totalScore:u,grade:c}}async generateBulletComments(e,t,n){let s=this.generateBulletCommentsPrompt(e,t,n),a=await this.callAPI(s);return this.parseAIResponse(a,"\u5F39\u5E55\u751F\u6210").map(c=>({...c,platform:t}))}async compareWithHit(e,t){let n=`\u8BF7\u5BF9\u6BD4\u5206\u6790\u4EE5\u4E0B\u4E24\u4E2A\u77ED\u5267\u5267\u672C\uFF1A

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

User-provided script content is transmitted to external model APIs, which is a genuine data-exposure concern when scripts may contain unpublished creative work or sensitive business material. The absence of an explicit warning or consent checkpoint at the transmission point makes accidental disclosure more likely in an agent setting.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

The skill name, description, and example invocation are entirely presented in Chinese, including the direct-use command example. There is no indication that the skill supports user language choice or that the Chinese-only presentation is a documented, justified locale restriction.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest says the skill supports platforms such as Douyin, Bilibili, Kuaishou, and Xiaohongshu, implying broad multi-platform support, but the code only hard-codes these four platform styles/enums. There is no extensible handling for other platforms beyond those exact values.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

The manifest advertises an "情绪热力图" capability, but the implementation formats simple percentage summaries like 好评/吐槽/疑问 rather than generating a structured heatmap or time/location-based emotional visualization. This is a semantic mismatch between the marketed capability and the actual behavior returned by the skill.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The manifest description is entirely in Chinese and does not indicate that the skill is region-specific or that users can opt into another language. Under the policy rule for natural-language violations, forcing a specific language without user choice can be a locale policy issue.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
90% confidence
Finding

Using a caret range for chalk allows npm to resolve newer releases than the exact version originally tested, which weakens build reproducibility and can silently introduce vulnerable or malicious upstream changes. In a CLI-distributed skill, supply-chain drift is a real risk because installs may pull different dependency trees over time.

Content

Scanner excerpt · package.json (reported line 32)May include surrounding context.

json
"author": "45号剧坊",
  "license": "MIT",
  "dependencies": {
    "chalk": "^5.3.0",
    "commander": "^12.1.0",
    "inquirer": "^9.2.20",
    "node-fetch": "^3.3.2",

Unverifiable Dependency: chalk has 1 known advisory(ies) (MAL-2025-46969 (Malicious code in chalk (npm))), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
77% confidence
Finding

The manifest uses an unpinned range for chalk while there is at least one known advisory associated with the package, making it impossible to verify from package.json whether installs resolve to a safe release. Combined with an executable CLI package, this uncertainty increases supply-chain risk because consumers may receive a compromised or vulnerable dependency version.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
90% confidence
Finding

Using a caret range for commander permits unreviewed minor/patch updates to be installed, reducing reproducibility and increasing exposure to supply-chain compromise or accidental breaking changes. For an executable package exposing a CLI entrypoint, this increases the chance that end users receive a dependency set different from what the publisher validated.

Content

Scanner excerpt · package.json (reported line 33)May include surrounding context.

json
"license": "MIT",
  "dependencies": {
    "chalk": "^5.3.0",
    "commander": "^12.1.0",
    "inquirer": "^9.2.20",
    "node-fetch": "^3.3.2",
    "node-html-parser": "^6.1.13",

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
90% confidence
Finding

Using a caret range for inquirer allows the installed dependency version to float, which can introduce unreviewed code changes and supply-chain risk. This is especially relevant for interactive CLI tooling because users commonly install directly from npm and trust the resolved package tree.

Content

Scanner excerpt · package.json (reported line 34)May include surrounding context.

json
"dependencies": {
    "chalk": "^5.3.0",
    "commander": "^12.1.0",
    "inquirer": "^9.2.20",
    "node-fetch": "^3.3.2",
    "node-html-parser": "^6.1.13",
    "ora": "^7.0.1",

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
93% confidence
Finding

Using a caret range for node-fetch reduces determinism and may allow installation of a release affected by known vulnerabilities or future malicious updates. Because this package likely performs network operations, dependency drift here is somewhat more sensitive than for purely cosmetic libraries.

Content

Scanner excerpt · package.json (reported line 35)May include surrounding context.

json
"chalk": "^5.3.0",
    "commander": "^12.1.0",
    "inquirer": "^9.2.20",
    "node-fetch": "^3.3.2",
    "node-html-parser": "^6.1.13",
    "ora": "^7.0.1",
    "yargs": "^17.7.2",

Unverifiable Dependency: node-fetch has 3 known advisory(ies) (CVE-2022-0235 (node-fetch forwards secure headers to untrusted sites); CVE-2022-2596 (node-fetch Inefficient Regular Expression Complexity ); CVE-2020-15168 (The `size` option isn't honored after following a redirect in node-fetch)), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
86% confidence
Finding

node-fetch has known advisories and the caret version range prevents confirming that all installations will use a patched release. Because node-fetch handles outbound HTTP behavior, affected versions can create real security issues such as header leakage, denial of service, or improper redirect handling when processing untrusted URLs or responses.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
88% confidence
Finding

Using a caret range for node-html-parser permits unpinned upgrades, which weakens reproducibility and increases supply-chain exposure. While not proof of compromise by itself, it is a valid dependency hygiene weakness that can make future vulnerable releases harder to control.

Content

Scanner excerpt · package.json (reported line 36)May include surrounding context.

json
"commander": "^12.1.0",
    "inquirer": "^9.2.20",
    "node-fetch": "^3.3.2",
    "node-html-parser": "^6.1.13",
    "ora": "^7.0.1",
    "yargs": "^17.7.2",
    "zod": "^3.22.4"

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
88% confidence
Finding

Using a caret range for ora allows upstream changes to be introduced without explicit review, creating a low-severity supply-chain and reproducibility issue. Even for non-security-critical UI libraries, floating versions can still bring malicious or vulnerable code into the runtime dependency graph.

Content

Scanner excerpt · package.json (reported line 37)May include surrounding context.

json
"inquirer": "^9.2.20",
    "node-fetch": "^3.3.2",
    "node-html-parser": "^6.1.13",
    "ora": "^7.0.1",
    "yargs": "^17.7.2",
    "zod": "^3.22.4"
  },