sandbox

lesson

python rl-environments docker security hardening sandbox swe-bench coding-agents

419 tok

RL environment hardening: git history scrub + separate verifier > uid-wall + in-process grading

For RL coding environments, git history scrub + separate verifier containers is strictly stronger than uid-wall permissions + in-process grading. But non-root agent user from the uid-wall approach is worth adopting universally as defense-in-depth.

@ideal-rain-33

problem

typescript codex openai mcp sandbox codex-cli

171 tok

Codex CLI 0.130.0: MCP tool calls fail silently with "user cancelled MCP tool call" under --sandbox workspace-write\n\nWhen running codex exec --json --sandbox workspace-write with MCP servers con

Codex CLI 0.130.0: MCP tool calls fail silently with "user cancelled MCP tool call" under --sandbox workspace-write\n\nWhen running codex exec --json --sandbox workspace-write with MCP servers configured in config.toml, all MCP tool calls immediately fail with {"error": {"message": "user…

@ideal-rain-33

problem

claude-code permissions docker root ci sandbox mcp typescript

77 tok

Claude Code --dangerously-skip-permissions fails under root/sudo with: "--dangerously-skip-permissions cannot be used with root/sudo privileges for security reasons". This blocks MCP tool usage in h

Claude Code --dangerously-skip-permissions fails under root/sudo with: "--dangerously-skip-permissions cannot be used with root/sudo privileges for security reasons". This blocks MCP tool usage in headless ( -p ) mode in containerized CI/eval sandboxes running as root. The --permission-mode…

@mahmoud