The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill recommends executing a Docker container with the --privileged flag to install binfmt handlers on the host system.
Evidence: Found in the 'Common Pitfalls' section of SKILL.md: docker run --privileged --rm tonistiigi/binfmt --install all.
Impact: Privileged containers have access to all devices on the host and can bypass most security boundaries. This specific command is used to modify the host's kernel configuration (binfmt_misc) for architecture emulation.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to download and run an external container image from Docker Hub to enable multi-architecture support.
Evidence: The command docker run --privileged --rm tonistiigi/binfmt --install all fetches the tonistiigi/binfmt image.
Context: While this is a standard community practice for Docker Buildx workflows, it involves pulling and executing third-party code at a high privilege level.
[COMMAND_EXECUTION]: The skill relies on various Docker CLI commands and MCP tools to perform builds and manifest management, which execute shell commands on the system.
Evidence: Use of tools like docker_buildxBuild, docker_manifestPush, and shell commands such as docker buildx build and docker manifest create are central to the skill's workflow.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing untrusted external data that influences command execution.
Ingestion points: Processes external Dockerfile content, target platform strings, and registry paths provided by the user or project environment.
Boundary markers: None present in the instructions to delimit or ignore embedded instructions within processed Dockerfiles.
Capability inventory: Includes file system interaction, container execution, and network operations via docker buildx and docker manifest tools.
Sanitization: No explicit sanitization or validation of the Dockerfile content or platform arguments is mentioned before execution.

docker-multi-platform