mcp-server-bash-sdk

Warn

Audited by Gen Agent Trust Hub on Jun 15, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs users to clone a repository from an unverified GitHub account (github.com/muthuishere/mcp-server-bash-sdk) which is not associated with the skill author or a recognized trusted vendor. This introduces a supply chain risk.
  • [COMMAND_EXECUTION]: Code examples for 'System Command Wrappers' use user-supplied input directly in shell commands (e.g., ps aux | grep -i "$filter"), creating a significant risk of command injection if the input contains shell-active characters.
  • [DATA_EXFILTRATION]: The provided 'File Operations' templates allow the agent to read and write files using user-provided paths. The implementation fullpath="${WORKSPACE_DIR}/${filepath}" lacks directory traversal protection (e.g., checking for ../), which could lead to unauthorized access to sensitive system files.
  • [PROMPT_INJECTION]: The skill architecture is vulnerable to indirect prompt injection because it passes untrusted data from an AI agent to bash tools without adequate sanitization.
  • Ingestion points: Tool functions such as tool_get_weather and tool_read_file ingest arguments directly via the $1 parameter in SKILL.md.
  • Boundary markers: No delimiters or explicit 'ignore embedded instructions' warnings are present in the provided templates to isolate adversarial input.
  • Capability inventory: The examples utilize powerful system utilities including curl, cat, ps, and shell redirection to files.
  • Sanitization: The implementation uses jq -r for data extraction but omits necessary shell escaping and path validation before executing system calls.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 15, 2026, 01:04 AM
Security Audit — agent-trust-hub — mcp-server-bash-sdk