Back to skill

Security audit

胡田 OPC导师 技术尽调 英文版

Security checks for vulnerabilities and agentic risk

Overview

This is a legitimate due-diligence skill, but it needs review because it can upload and persist confidential deal materials in an external IMA knowledge base without clear consent, retention, or destination controls.

Review this before installing for confidential engagements. Use it only with approved IMA tenants and scoped API keys, require explicit customer consent before importing files or writing reports back, confirm the exact destination document or knowledge base, and disable or sanitize web searches for sensitive projects.

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
references/Coze-IMA协同尽调流程.md:126
Finding

Uncontrolled External Upload and Persistence of Confidential Due-Diligence Data

Content
View full analysis

Vulnerability Details

File Location: references/Coze-IMA协同尽调流程.md, lines 126–140 and 359–369
Vulnerability Type: Uncontrolled external disclosure and storage of sensitive data
Risk Level: Medium

Complete Vulnerable Snippets

Lines 126–140 instruct the agent to upload customer documents:

markdown
#### 1.4 New Material Import

If the customer provides new documents, use the `import_doc` API:

API: import_doc
Parameters:
  - file: document file (PDF/DOCX/TXT)
  - title: document title (recommended format: project_name_material_type_date)

Import priority:
1. Business plan
2. Technical white paper/product specification
3. Patent documents
4. Team introduction/resumes
5. Financial data/test reports

Lines 359–369 instruct the agent to persist the completed report:

markdown
#### 4.4 Write the Report Back to IMA

Use the `append_doc` API to write the final report into the IMA knowledge base:

API: append_doc
Parameters:
  - doc_id: target document ID
  - content: report content (Markdown format)

Recommended naming format:
Due_Diligence_Report_ProjectName_YYYYMMDD

The original source is written in Chinese; the snippets above are faithful English translations so that all report text complies with the required language.

Technical Analysis

The workflow handles business plans, technical specifications, patent documents, resumes, financial information, and test reports. These materials may contain trade secrets, personal data, financial data, and other confidential information.

The instructions direct the agent to transmit those files to an external IMA service and later persist an aggregated due-diligence report. They do not require:

  • Explicit and informed customer authorization before transmission.
  • Data classification or sensitivity review.
  • Redaction or minimization of personal, financial, or proprietary data.
  • Verification of the destination tenant, account, knowledge base, or document.
  • A local-only processing alter ...[truncated 1924 chars]
Remediation
View remediation

Remediation Suggestions

  1. Require explicit authorization

    • Present a clear notice identifying the external service, destination tenant, categories of data, and purpose.
    • Require affirmative customer approval before each upload or report write-back.
    • Do not treat general use of the skill as consent to external storage.
  2. Default to local-only processing

    • Make external import and persistence opt-in.
    • Provide a mode that analyzes supplied files without transferring or retaining them in a third-party repository.
  3. Verify the destination

    • Display and confirm the IMA account, tenant, knowledge-base ID, and target document ID.
    • Use an allowlist of approved destinations.
    • Reject unknown, personal, or cross-customer repositories.
  4. Minimize and redact data

    • Upload only files and excerpts necessary for the requested analysis.
    • Detect and redact credentials, private keys, personal identifiers, bank information, confidential contact details, and unrelated records.
    • Never upload .env files, secret stores, credential files, or unrelated local files.
  5. Apply retention controls

    • Define a retention period before upload.
    • Support immediate deletion after analysis.
    • Record deletion status and provide the customer with destination and retention details.
  6. Harden access control

    • Use dedicated, least-privilege service credentials.
    • Limit credentials to the approved knowledge base and required API operations.
    • Avoid broad inspection of secret files or environment variables.
    • Keep audit logs of uploads, write-back actions, destinations, and authorization decisions.
  7. Add mandatory preflight checks

    • Classify each document before transmission.
    • Block uploads involving regulated or highly confidential data unless the destination is explicitly approved.
    • Verify encryption in transit, encryption at rest, tenant isolation, and access policy.
  8. **Prevent unintended p ...[truncated 179 chars]

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (12)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The title says '英文版' (English version), but the actual user-facing content is written in Chinese. This creates a language/locale policy concern because the skill content effectively forces a specific language without offering user choice or documenting a justified locale restriction.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The manifest description is entirely in Chinese and the rest of the skill instructions are also written only in Chinese, which effectively imposes a language/locale on users. There is no opt-in, alternative language option, or note that the skill is intended only for a Chinese-speaking or region-specific compliance context.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The workflow tells operators to import customer documents into IMA without clearly warning that uploaded files will be stored in an external knowledge base. For technical due diligence, those documents may include proprietary technical, financial, patent, or team information, so lack of transparent consent can lead to unintended disclosure and retention of sensitive client data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The process instructs the agent to fall back to web search when IMA lacks results, but it does not warn that project names, founder identities, patent numbers, or technical keywords may be disclosed to external search providers. In a due-diligence setting, even the existence of a query can be commercially sensitive and may leak confidential deal activity or target information.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest describes a tool for technical feasibility assessment, team background checks, patent verification, and business-model analysis. Lines L339-L344 add a separate capability to produce channel-fit analysis and channel strategy recommendations, which is not part of the stated due-diligence screening scope and shifts the skill toward business development planning.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The workflow instructs the agent to append final due-diligence reports back into the IMA knowledge base, creating persistent storage of potentially sensitive client materials, findings, and risk conclusions. In a due-diligence context, these reports may contain confidential business, legal, and reputational information, so silent write-back materially increases data-retention and unauthorized-access risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

Appending the final due-diligence report into IMA persists sensitive conclusions, red flags, and potentially defamatory or legally sensitive assessments without an explicit warning about storage and downstream access. Because due-diligence reports often combine internal analysis with imported source material, silent persistence can magnify confidentiality and compliance risks.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The document explicitly directs the process to source authentication material from SECRET.md or environment variables, introducing secret-access behavior that exceeds what a read-only due-diligence description needs to expose. In an adversarial or over-permissioned agent setting, normalizing access to secrets can enable credential misuse, unauthorized API operations, and lateral access into the IMA environment.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The entire document is written in Chinese and provides no indication that other languages are supported or that the user can opt into this locale. Under the stated policy, forcing a specific language without user choice can be a natural-language policy violation unless the locale constraint is explicitly documented and justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

This markdown file is entirely in Chinese, beginning with the Chinese title and continuing throughout, with no indication that the language requirement is optional or limited to a China-specific audience. Under the policy rule for natural-language violations, forcing a specific language without user opt-in can be a locale-policy issue.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

SQP-3 applies to all file types and covers language or locale policy violations. This markdown file presents all instructions and analysis content only in Chinese, and nowhere indicates that the user can choose another language or that the Chinese-only format is a documented, justified regional constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

This markdown file uses Chinese throughout and does not indicate that the user can choose another language or that the content is intentionally limited to a Chinese-speaking audience. Under the stated policy, forcing a specific language without user opt-in can be a natural-language policy violation.

Content

No source excerpt is available for this finding.