Back to skill

Security audit

视频创作

Security checks for vulnerabilities and agentic risk

Overview

This skill is only a placeholder, but it asks for unrestricted shell access without explaining or limiting what it will run.

Review this carefully before installing. The package does not contain the actual video-creation workflow, and its unrestricted Bash permission is broader than the placeholder justifies. Only install after reviewing the complete referenced repository content and narrowing or removing shell access unless a specific, documented command set is required.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:4
Finding

Unrestricted Bash Permission Violates Least Privilege

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 4
Vulnerability Type: Excessive tool permissions
Risk Level: Medium

Vulnerable Code

yaml
allowed-tools: Bash(*)

Technical Analysis

The skill declares unrestricted access to the Bash tool through the wildcard permission Bash(*). However, the package contains only a placeholder document and defines no local operation that legitimately requires shell execution.

This permission is broader than the skill's documented functionality and violates the principle of least privilege. If the agent runtime grants the declared capability, instructions added later or obtained from the referenced external repository could invoke arbitrary shell commands without being constrained to a narrowly defined command set.

The external repository reference does not itself download or execute content, so remote payload retrieval and execution is not confirmed by the audited package.

Attack Path

  1. A user loads or enables the skill.
  2. The agent runtime grants the declared wildcard Bash capability.
  3. Untrusted, externally obtained, or subsequently modified skill instructions direct the agent to invoke Bash.
  4. Because the permission has no command-level restriction, the instructions can attempt arbitrary commands under the privileges and sandbox boundaries of the agent process.
  5. Those commands may access or alter resources available to that process.

Impact Assessment

Successful exploitation could permit arbitrary command execution with the effective privileges of the agent process. Depending on runtime sandboxing, operating-system permissions, and approval controls, the accessible scope could include project files, environment variables, local credentials, network resources, and other files writable by the executing account.

The audited file contains no command that actively exercises this permission. Therefore, the finding represents an excessi ...[truncated 91 chars]

Remediation
View remediation

Remediation Suggestions

Remove allowed-tools: Bash(*) because the current placeholder functionality does not require shell access.

If executable functionality is added later:

  1. Include all required implementation files in the reviewed package rather than relying on unspecified external instructions.
  2. Grant only the exact tools and narrowly scoped commands required for the documented task.
  3. Avoid wildcard shell permissions and require explicit approval for commands involving the network, credentials, package installation, or destructive file operations.
  4. Validate and safely quote all command arguments derived from users or external data.
  5. Run commands in a restricted sandbox using a low-privilege account with minimal filesystem and network access.
  6. Re-audit the complete downloaded implementation before enabling any execution permission.
Vulnerability Patterns
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (4)

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

Unrestricted Bash access is unjustified by the stated purpose, which is merely a cloud-synced placeholder. Because there are no visible controls, task boundaries, or command restrictions, this permission could be abused to run arbitrary local commands, access files, exfiltrate data, or modify the environment once the skill is completed or invoked.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest grants unrestricted Bash execution even though the file is only a placeholder that instructs users to fetch content from a remote repository later. This creates a dangerous trust gap: a minimally documented skill can obtain full shell capability before its real behavior is reviewed, enabling arbitrary command execution if additional synced content is later loaded or the skill is activated unexpectedly.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The manifest description only states that this is a skill synced from a cloud repository, but provides no concrete invocation phrases, scope limits, or exclusion conditions. For a manifest/markdown file, this absence of specificity can make it unclear when the skill should activate versus when it should not.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The natural-language content in the description and instructions is entirely in Chinese, with no indication that the user can choose another language or that the skill is intentionally region-specific. This can violate language/locale policy when a skill effectively constrains communication to a specific language without opt-in.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.