Show HN: ActraDeck – Put risky coding-agent actions back in front of a human (github.com)

ActraDeck has emerged as a promising new tool aimed at enhancing developer safety by allowing human oversight for risky actions performed by AI coding agents like Claude Code and Codex. This local cockpit flags high-risk actions, such as a recursive filesystem delete command, and requires user approval to proceed. Notably, ActraDeck operates entirely on local machines without requiring a cloud account, making it readily accessible for developers while providing features such as masking detected secrets and session replay capabilities. The significance of ActraDeck lies in its focus on preventing reckless approvals of potentially harmful commands, enabling developers to maintain a more secure coding environment. By presenting high-risk requests in an approval inbox, it not only supports immediate decision-making but also provides a valuable audit trail of actions and responses. Though still in early development, ActraDeck promises essential features like event redaction, local audits, and a risk classification system, positioning it as a vital addition for developers, platform engineers, and teams managing multiple coding agents while prioritizing security.