fix: sanitize subprocess call in init_agent.py

[orbisai0security]

Summary

Fix critical severity security issue in skills/agent-builder/scripts/init_agent.py.

Vulnerability

Field	Value
ID	V-001
Severity	CRITICAL
Scanner	multi_agent_ai
Rule	V-001
File	skills/agent-builder/scripts/init_agent.py:66
CWE	CWE-78

Description: The agent builder script passes user-controlled command strings directly to subprocess.run() with shell=True at two locations (lines 66 and 136). The command values originate from agent or user input (b.input["command"] and args["command"]) and are not sanitized, validated, or restricted before execution. When shell=True is used, the operating system shell interprets the entire string, enabling injection of shell metacharacters such as semicolons, pipes, backticks, and command substitution sequences. An attacker who can influence the command string — either directly via CLI input or by crafting a prompt that causes the LLM to invoke the command tool with a malicious payload — can execute arbitrary OS commands.

Changes

• skills/agent-builder/scripts/init_agent.py

Verification

• Build passes
• Scanner re-scan confirms fix
• LLM code review passed

---

Automated security fix by OrbisAI Security

[vercel]

@orbisai0security is attempting to deploy a commit to the crazyboym's projects Team on Vercel.

A member of the Team first needs to authorize it.