The Agent Skills Directory

[COMMAND_EXECUTION]: The skill allows the execution of arbitrary commands within a Docker container to support development workflows.
Evidence: The instructions utilize docker exec "$ARENA_CONTAINER" su $(id -un) -c "... <command>" to run commands. Notably, it drops root privileges by switching to the host user's ID within the container.
[COMMAND_EXECUTION]: The skill executes local repository scripts to manage the lifecycle of Docker images.
Evidence: Execution of ./docker/run_docker.sh -r and -R from the repository root to rebuild images.
[INDIRECT_PROMPT_INJECTION]: The skill provides an execution surface where user-supplied instructions are interpolated into shell commands inside a container.
Ingestion points: Development commands provided by the user or derived from repository state are passed to the docker exec template in SKILL.md.
Boundary markers: No explicit delimiters are used for the command string.
Capability inventory: The skill can execute shell commands via Docker and run local scripts (run_docker.sh).
Sanitization: No explicit sanitization or filtering is performed on the command string before execution within the container.

dev-container