Back to skill

Security audit

vibegit

Security checks for vulnerabilities and agentic risk

Overview

ClawScan’s verdict is benign with medium confidence; AIG reported no findings, while Skillspector flagged mismatches between the skill’s stated read-only, no-network purpose and commands reported to install or clone code.

The verdict is benign, but the medium confidence and Skillspector’s reported mismatch are worth considering before installing. Its analysis reports complete file coverage; AIG reported no findings. The supplied evidence does not verify the behavior of the downloaded script beyond its help command.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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 (7)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied commands explicitly download and write a full repository, contradicting the declared no-network and read-only restrictions. Invoking vibegit.py --help also executes downloaded code; the script's internal behavior is not supplied, so its analysis functionality or additional effects cannot be verified.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied shell commands explicitly perform a network-backed git clone, which writes repository data and checks out files locally. This contradicts the declared read-only, commit-metadata-only, no-network scope. Running --help may support setup, but that does not negate the explicit resource and access restrictions. The chunk does not establish what the Python script itself does.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The only supplied command installs a skill; it does not implement the declared metadata-only assessment. Installation materially differs from read-only analysis and ordinarily retrieves and writes package content. The exact network activity and installed skill behavior cannot be established from this snippet, but the installation operation itself contradicts the declared read-only purpose.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The visible command performs installation, not the advertised analysis. Although installation could be a setup step, it is materially inconsistent with the explicit read-only and no-network boundaries. The declared file-write permission does not remove that descriptive contradiction. The supplied chunk does not establish what the installed package subsequently does.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The only supplied command is an installation operation, not an implementation of the described metadata-only audit. Installation is materially different from read-only repository inspection and ordinarily writes local installation data and may contact a package registry. The exact network and filesystem effects cannot be confirmed from this command alone.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · vibegit.py (reported line 66)May include surrounding context.

python
# ---------------------------------------------------------------- git access
def run_git(repo, args, timeout=180):
    proc = subprocess.run(
        ["git", "-C", repo] + list(args),
        stdout=subprocess.PIPE, stderr=subprocess.PIPE,
        text=True, encoding="utf-8", errors="replace", timeout=timeout,

Permission 'file_read' is declared but no corresponding code capability (file_read) was detected.

Low
Category
MCP Least Privilege
Confidence
65% confidence
Finding

Declared permissions with no matching code capability may indicate removed functionality or pre-staging for future abuse.

Content

No source excerpt is available for this finding.