create-mcp-skill
Warn
Audited by Gen Agent Trust Hub on Feb 23, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- EXTERNAL_DOWNLOADS (MEDIUM): The skill recommends installing the
mcpCLI from an untrusted GitHub account (github.com/f/mcptools). Installing binaries from unverified third parties is a security risk.\n- REMOTE_CODE_EXECUTION (MEDIUM): The documentation promotes usingnpxandbunxto run packages from remote registries at runtime. This practice lacks version pinning and integrity verification, exposing the environment to supply chain vulnerabilities.\n- COMMAND_EXECUTION (LOW): The use ofpkill -9 -ffor process management is aggressive. If an attacker controls the process name pattern, they could potentially terminate unrelated system processes.\n- PROMPT_INJECTION (LOW): The 'Template Generator' functionality allows for indirect prompt injection by interpolating untrusted variables into a shell command that writes to the filesystem.\n - Ingestion points: $SKILL_NAME and $SERVER_COMMAND variables.\n
- Boundary markers: Absent.\n
- Capability inventory: Write (cat), Bash (Command Execution).\n
- Sanitization: Absent.
Audit Metadata