The rise of AI agents like Codex, Claude Code, and Cline has brought powerful automation to development workflows. These tools can read project files, execute shell commands, call MCPs, and access the network, sometimes running for extended periods. However, this capability introduces a significant security challenge: how do you prevent an agent from performing harmful actions? KolaCapa, an open-source project, proposes a solution by using YAML configuration files to define explicit permission boundaries. This approach allows developers to specify which commands an agent can run, which files it can access, and which network endpoints it can reach. By treating agents as privileged processes with auditable policies, teams can integrate AI assistance without sacrificing security. This pattern is likely to become a standard practice as agent adoption grows, making permission management a key consideration for any organization deploying AI tools.
AI agents can now run commands and access files, but who controls their permissions? This signal explores KolaCapa, an open-source project that uses YAML to define clear boundaries for agent actions, addressing a critical security gap.