Back to skill

Security audit

Framework Analysis

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly an analysis workflow, but it automatically runs an unreviewed local converter and sends generated reports to Feishu without a clear approval step.

Review this skill before installing. It may be useful for Chinese-language framework analysis, but only use it if you are comfortable with automatic research delegation, local report generation, execution of an external converter script, and Feishu transmission. Prefer requiring explicit confirmation of the output path, recipient, and file contents before any send action.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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

Error
Location
SKILL.md:143
Finding

Automatic External Transmission Without User Authorization

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 143–149
Vulnerability Type: Automatic transmission of generated files to an external messaging service
Risk Level: High

Vulnerable code:

text
### Step 2: 发送飞书

使用 message tool 发送 PDF:
javascript
message({ action: "send", message: "📊 [主题]框架法终局推导", filePath: "research/[topic]-analysis-[date].pdf" })

Technical Analysis

The Skill instructs the agent to transmit every generated report through the Feishu messaging integration automatically. It does not require explicit user consent, display or validate the destination, classify the report's sensitivity, or provide a local-only delivery option.

Because the report is derived from user input and external research, it may contain confidential topics, sensitive assessments, personal information, or other material that the user did not intend to disclose outside the current environment. The instruction crosses a trust boundary from local analysis and file generation to an external service without an authorization checkpoint.

Attack Path

  1. A user requests a framework analysis and supplies a potentially sensitive subject or supporting information.
  2. The Skill incorporates the supplied information and research results into a Markdown report.
  3. The report is converted into a PDF.
  4. The Skill automatically invokes the message tool with the generated PDF.
  5. The PDF is uploaded to the Feishu destination associated with the active tool context without confirming the recipient or obtaining the user's approval.
  6. Anyone with access to that destination may obtain the report and its sensitive contents.

Impact Assessment

Exploitation does not grant new operating-system privileges, but it abuses an already authorized messaging capability beyond the minimum permissions needed to perform local analysis. The exposed scope can include the entire generated report, user-supplied info ...[truncated 290 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove automatic external delivery and make local report generation the default behavior.
  2. Require explicit, per-transmission user confirmation after the report has been generated.
  3. Before confirmation, display the exact file path, destination service, channel or recipient, and a concise description of the data being sent.
  4. Validate the destination against an approved allowlist rather than relying on ambient messaging context.
  5. Add a sensitivity review that detects credentials, personal information, confidential input, and restricted material before upload.
  6. Apply least privilege to the messaging integration and prevent the Skill from choosing arbitrary recipients.
  7. Record an audit event for each approved transmission without logging the report's sensitive contents.
  8. Provide a local-only or manual-download alternative when authorization is absent.

T08 · Insecure Dependencies

Warning
Location
SKILL.md:131
Finding

Execution of an Unbundled and Unverified Cross-Skill Script

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 131–139
Vulnerability Type: Mutable external component executed without integrity verification
Risk Level: Medium

Vulnerable code:

bash
~/.openclaw/skills/deep-research-zh/scripts/md2pdf.sh research/[topic]-analysis-[date].md research/[topic]-analysis-[date].pdf

Technical Analysis

The Skill directs the agent to execute md2pdf.sh from a different Skill directory. That script is not included in the audited project, so its implementation, ownership, update policy, and integrity cannot be verified from this artifact.

The absolute reference does not pin the script to a reviewed version or verify a cryptographic digest before execution. Consequently, the effective behavior of this Skill can change after review if another package installation, update, or local modification replaces the referenced script.

This finding does not establish that the referenced script is currently malicious. The vulnerability is the unconditional execution of a mutable, unaudited cross-Skill dependency under the agent's existing permissions.

Attack Path

  1. An attacker or compromised installation process gains the ability to modify ~/.openclaw/skills/deep-research-zh/scripts/md2pdf.sh.
  2. The attacker replaces or alters the script to include additional commands while preserving expected PDF-generation behavior.
  3. A user invokes the framework-analysis Skill.
  4. During automatic delivery, the Skill executes the modified script.
  5. The injected commands run with the same operating-system permissions and environment access as the agent process.
  6. Depending on those permissions, the script can read accessible files, alter generated reports, invoke network utilities, or execute other local commands.

Impact Assessment

Successful exploitation provides command execution with the privileges of the agent process; it does not inherently elevate privileges abo ...[truncated 392 chars]

Remediation
View remediation

Remediation Suggestions

  1. Bundle the conversion implementation within this Skill so that it is included in the same review and release lifecycle.
  2. Pin the converter to a reviewed version and verify a cryptographic hash or trusted signature before every execution.
  3. Validate that the script is a regular file owned by an expected account and is not writable by untrusted users.
  4. Avoid mutable cross-Skill paths under the user's home directory.
  5. Execute document conversion in a sandbox with restricted filesystem access, no unnecessary environment variables, and no network access.
  6. Pass input and output paths as safely quoted arguments and constrain both paths to a dedicated working directory.
  7. Require user confirmation before executing local conversion commands.
  8. Fail safely if dependency verification fails; do not execute an unverified fallback.
  9. Audit the converter and its underlying PDF engines independently, including their handling of untrusted Markdown, HTML, local-file references, and external resources.
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 (7)

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill instructs execution of a shell command that invokes an external script to transform analysis output into a PDF. Shell/script execution is a materially higher-risk capability than text analysis because it can create files, touch the filesystem, and become a vehicle for command/path injection or unsafe script behavior if topic-derived paths are not strictly controlled.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill automatically sends the generated analysis file through a messaging tool, creating an outbound exfiltration path. Even if intended for convenience, automatic transmission of potentially sensitive or policy-relevant content without a hard approval step can leak data to external recipients or channels.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

Automatically sending the generated analysis via Feishu without a clear warning or consent introduces a direct data-leakage risk. Because the content concerns geopolitical or historical analysis, outputs may include sensitive summaries, and silent transmission makes accidental disclosure significantly more dangerous.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
80% confidence
Finding

The natural-language description and operational instructions are entirely in Chinese and indicate a Chinese-oriented workflow, but the file does not offer users a language choice or explain that the skill is intentionally restricted to a Chinese-language context. This can violate language/locale policy when a specific language is effectively forced without user opt-in.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

Automatically spawning parallel sub-agents causes the skill to initiate additional research actions beyond a single, bounded analysis flow. This can widen data access, increase unreviewed tool usage, and make it harder for users to understand or control what external retrieval is being performed.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill’s declared purpose is analytical reasoning, but the instructions add post-analysis operational behavior: local file generation and outbound delivery. That expands the trust boundary from reasoning into side-effecting actions, increasing the chance of unintended data handling, disclosure, or abuse if sensitive analysis is produced.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill describes automatic PDF generation and file creation without clearly warning the user or obtaining consent. Silent creation of artifacts on disk is risky because it may store sensitive analysis in predictable locations, surprise users, or violate least astonishment and operational safety expectations.

Content

No source excerpt is available for this finding.