Back to skill

Security audit

zeelin-text-analysis

Security checks for vulnerabilities and agentic risk

Overview

This skill is a cloud-backed text analysis workflow, but it sends user documents and an App-Key over a plaintext HTTP endpoint and leaves important privacy and security controls unclear.

Review before installing. Only use this skill for documents you are willing to send to the Zeelin service and OSS storage, do not submit confidential material unless the provider documents acceptable security and retention terms, and avoid entering a real App-Key until the service uses HTTPS with a verified domain. Rotate any App-Key previously used through the plaintext endpoint.

Vulnerability Patterns
  • 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
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (2)

T09 · Insecure Skill Coding Practices

Error
Location
templates/config.json:4
Finding

Plaintext Transmission of Documents, Credentials, and Analysis Results

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:67-69, SKILL.md:98-104, SKILL.md:125, SKILL.md:158, and templates/config.json:4
Vulnerability Type: Sensitive data transmitted over unencrypted HTTP
Risk Level: High

Vulnerable Code

templates/config.json:1-6

json
{
  "skill_name": "Text Analysis Skill",
  "skill_version": "1.0.0",
  "service_url": "http://47.98.180.113:8083",
  "Zeelin_App_Key": "YOUR_REAL_APP_KEY",
  "description": "Text entity and relationship extraction with local file upload support"
}

SKILL.md:67-69

bash
curl -X POST "http://47.98.180.113:8083/api/skill/upload" \
  -F "appKey=YOUR_APP_KEY" \
  -F "file=@/path/to/local/file.txt"

SKILL.md:98-104

bash
curl -X POST "http://47.98.180.113:8083/api/skill/text" \
  -H "Content-Type: application/json" \
  -d '{
    "appKey": "YOUR_APP_KEY",
    "oss": "https://jumuai.oss-cn-hangzhou.aliyuncs.com/..."
  }'

SKILL.md:158

bash
curl "http://47.98.180.113:8083/api/skill/status/analysis_xxx"

Technical Analysis

The configured API endpoint uses unencrypted HTTP and a raw IP address. The documented workflow transmits the user's App-Key and the complete contents of a selected document to this endpoint. It subsequently sends an externally hosted object URL and retrieves task status and analysis results through the same plaintext transport.

HTTP provides neither confidentiality nor authenticated transport integrity. An attacker with visibility or control over the network path can inspect requests, capture credentials and document contents, alter submitted data, or modify task-status responses. The HTTPS website referenced elsewhere in the documentation does not protect traffic sent to the separate HTTP API endpoint.

Attack Path

  1. A user configures a valid App-Key and asks the Skill to analyze a document.
  2. The Agent uploads the App-Key and document to ` ...[truncated 1259 chars]
Remediation
View remediation

Remediation Suggestions

  1. Replace the HTTP endpoint with an HTTPS endpoint hosted on an authenticated domain and protected by a valid, publicly trusted certificate.
  2. Reject configuration values using http://; do not silently downgrade to plaintext transport.
  3. Ensure the client performs normal TLS certificate and hostname validation. Do not disable verification.
  4. Replace long-lived App-Keys with short-lived, revocable, least-privilege tokens where supported.
  5. Send credentials through a standard authorization header rather than embedding them in multipart forms or JSON bodies, while still requiring TLS.
  6. Obtain explicit user confirmation before uploading files and clearly identify the external recipient, transmitted data, retention policy, and billing implications.
  7. Apply server-side authorization to every upload, submission, and status request. Ensure task identifiers alone cannot authorize access to results.
  8. Use expiring, narrowly scoped object-storage URLs and avoid publicly readable OSS objects.
  9. Rotate any App-Key that may already have been transmitted through the plaintext endpoint.
  10. Add automated tests or configuration validation that fail when any production API URL uses unencrypted HTTP.

T08 · Insecure Dependencies

Warning
Location
SKILL.md:4
Finding

Unpinned and Unnecessary Third-Party Python Dependencies

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:1-5
Vulnerability Type: Unpinned supply-chain dependencies
Risk Level: Medium

Vulnerable Code

SKILL.md:1-5

yaml
---
name: zeelin-text-analysis-skill
description: "Text analysis skill supporting direct upload and analysis of local files."
metadata: { "openclaw": { "emoji": "📄", "requires": { "bins": ["python3"], "pip": ["flask", "requests", "oss2"] } } }
---

Technical Analysis

The Skill declares flask, requests, and oss2 without exact versions or integrity hashes. Dependency resolution can therefore select different package releases over time, including future releases that have not been reviewed with this Skill.

Package installation and import hooks may execute code with the permissions of the process installing or running the Skill. Consequently, compromise of a declared package, its dependency chain, or the configured package source could introduce code execution into the local Agent environment.

The audited package contains only documentation and configuration files; no included executable code was found that uses these libraries. Declaring apparently unused dependencies unnecessarily increases the supply-chain attack surface. The audit did not establish that any of the named packages are currently malicious.

Attack Path

  1. The Skill platform reads the pip requirements from the metadata.
  2. It resolves the unpinned package names against the configured Python package index.
  3. A compromised or unexpectedly changed current release, or a compromised transitive dependency, is selected because no reviewed version or hash is enforced.
  4. The package is installed or loaded in the Agent environment.
  5. Malicious installation or runtime code executes with the privileges available to the package manager or Agent process.
  6. Depending on those process permissions, the payload could access files, environment variables, networ ...[truncated 723 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove flask, requests, and oss2 unless the Skill contains reviewed code that demonstrably requires them.
  2. Pin every required direct dependency to an exact, reviewed version.
  3. Lock all transitive dependencies and verify package integrity with cryptographic hashes, such as a hash-locked requirements file.
  4. Retrieve packages only from an approved and authenticated package repository.
  5. Run automated vulnerability and license scanning against the complete resolved dependency graph.
  6. Review and deliberately approve dependency updates rather than accepting the latest available release automatically.
  7. Install dependencies in an isolated, non-privileged environment with restricted filesystem and network access.
  8. Prevent package installation scripts from receiving unnecessary secrets or access to user documents.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (15)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill instructs the agent to upload local user files and an App-Key to external services but does not provide a clear privacy, third-party transfer, or retention warning. This is dangerous because users may unknowingly send sensitive documents and credentials off-platform to remote infrastructure, including OSS-backed storage and a Zeelin service.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill is advertised as supporting direct local text analysis, but the documented workflow requires uploading the user's file to remote OSS and then submitting it to an external Zeelin service. This creates a material privacy and trust boundary mismatch: users may believe data stays local when in fact sensitive document contents are transmitted off-device to third parties.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger phrases are broad generic requests like '分析文本' and '分析这个文档', which can cause the skill to activate for many ordinary document interactions. In this skill's context, overbroad triggering is more dangerous because activation leads to external upload of user files and credential use, increasing the chance of accidental exfiltration.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
98% confidence
Finding

The documented workflow sends user content and an App-Key to an external network endpoint for processing, which is a clear external transmission path. In the context of a text-analysis skill handling local files, this is dangerous because it can expose confidential document contents and credentials to third-party systems without strong disclosure, minimization, or trust guarantees.

Content

Scanner excerpt · SKILL.md (reported line 97)May include surrounding context.

示例请求:

bash
curl -X POST "http://47.98.180.113:8083/api/skill/text" \
  -H "Content-Type: application/json" \
  -d '{
    "appKey": "YOUR_APP_KEY",

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The instruction to generate an md file from analysis results introduces local persistence of potentially sensitive content without telling the user that data will be written to disk. This can leave residual copies of analyzed material on the local system and expand the exposure surface beyond the original upload.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The result-formatting section mandates creating a Markdown file but omits any warning that the analyzed content may be persisted locally. Because the output may contain extracted entities, relationships, and character profiles from user documents, silent file creation can expose sensitive derived data to other local users, backups, or sync services.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The description says the skill supports direct local file upload and broad text-content analysis without clearly constraining file types, trust boundaries, or invocation conditions. In an agent setting, overly broad upload-analysis metadata can cause the skill to be invoked on sensitive local documents more readily than intended, increasing the risk of unnecessary exposure or processing of private data.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The config routes allegedly local text/file analysis to a remote HTTP endpoint, which means uploaded document contents may be transmitted off-device despite the skill description implying local handling. Because the endpoint uses plain HTTP, data in transit can be intercepted or modified, creating confidentiality and integrity risks for potentially sensitive uploaded text.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The stated purpose is local text-upload analysis, but the implementation points to an external network service without justification or user-visible disclosure. In this skill context, users are likely to upload documents that may contain sensitive business or personal information, so undisclosed exfiltration to a third-party service materially increases privacy and compliance risk.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The skill's stated purpose is analyzing text and presenting results, but multiple instructions tell the agent to generate an md file for the user to view. Creating output files is a separate file-generation capability that is not declared in the manifest's purpose statement. While related to presentation, it is not an obvious requirement of text analysis itself.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

This instruction goes beyond formatting results as Markdown and explicitly requires generating an md file for user viewing. File creation is an additional capability not stated in the manifest, which only describes text analysis of uploaded documents. The mismatch is small but still reflects undeclared behavior.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The documented complete workflow ends with the agent generating an md file to display results. That is a distinct output-generation behavior not reflected in the manifest, which frames the skill as a text-analysis tool for uploaded files. Users reading the manifest would not expect file creation as part of the skill's operation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill name and description are presented only in Chinese, with no indication that language selection is user-configurable or intentionally limited to a Chinese-speaking context. This can violate language/locale policy expectations when a skill implicitly forces a specific language without opt-in or documented regional justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The skill name and description are entirely in Chinese, with no indication that language selection is optional or that the skill is intended only for a Chinese-speaking context. That can create a language-policy issue if users are not given a locale choice or justification for the restriction.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

This JSON manifest describes the skill only as '文本实体关系提取,支持本地文件上传' without any explicit trigger phrases, scope boundaries, or exclusion conditions. In a manifest file, that lack of specificity can make activation overly broad or ambiguous, especially around generic text-analysis requests.

Content

No source excerpt is available for this finding.