codebootstrap

AI tools (`claude`, `codex`, `gemini`) are configured to run in 'YOLO mode' (`--dangerously-skip-permissions`, `--dangerously-bypass-approvals-and-sandbox`, `--yolo`) by default. This grants them significant autonomy without explicit user approval for actions such as file system modifications, Git operations, and API calls. While guidelines are provided to the AI within project context files (`context-append.md`) to prevent destructive operations, the bypass of explicit human confirmation poses a significant risk for unintended or malicious actions if the AI hallucinates or is prompted incorrectly. The `host-setup.sh` script is designed for `curl | bash` execution, which introduces a supply chain security risk by running unreviewed remote code. The `filesystem`, `git`, and `github` MCP servers grant broad access to the workspace filesystem and GitHub API. `GITHUB_TOKEN` is exposed to the `github` MCP server for API access.