The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The Dockerfile and toolchain reference templates include commands to download and execute installation scripts from well-known vendor domains.
The code-server installer is fetched from https://code-server.dev/install.sh and piped to the shell for installation.
Node.js environment setup scripts are fetched from NodeSource and executed during the image build process.
[PRIVILEGE_ESCALATION]: The skill configures high-level permissions within the container sandbox and provides an opt-in mechanism for host-level control.
The Dockerfile template grants passwordless sudo privileges to the unprivileged dev user within the container to allow for dynamic package installation and Docker socket management.
The docker-compose.yml template contains a commented-out entry for mounting the host's Docker socket (/var/run/docker.sock). While this grants root-equivalent access to the host machine, the skill explicitly documents this risk in references/security.md and labels it as an opt-in configuration for specific use cases like sibling container management.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from repository manifests, CI configurations, and scripts to generate environment configurations.
Ingestion points: The agent recursively scans repository files, including .github/workflows/, package.json, and other manifests across the project structure (references/detection.md).
Boundary markers: The skill instructions emphasize a strict security model and point the agent toward the references/security.md guide to prevent accidental credential leakage or unsafe configuration generation.
Capability inventory: The skill possesses the ability to read the full repository, write configuration files to the host (in the devbox/ folder), and execute commands via Docker.
Sanitization: The agent is instructed to perform a 'portability sweep' to identify and fix host-specific or problematic shell commands within the project's own scripts during the setup process.
[COMMAND_EXECUTION]: The skill workflow involves executing shell commands to build containers and verify tool installations as part of its verification step.
The agent runs docker compose build and uses docker exec to check tool versions and run project-specific tests to confirm the environment is functional.
[DYNAMIC_EXECUTION]: The container entrypoint allows for the execution of arbitrary commands defined in the DEVBOX_BOOTSTRAP environment variable, which is populated based on detected repository needs.
The assets/entrypoint.sh script executes bash -lc "${DEVBOX_BOOTSTRAP}" during the container startup process to handle project-specific initialization like dependency restoration.

devbox-anywhere