The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute multiple shell commands to perform environment discovery and container management.
Evidence in SKILL.md: docker --version, docker info, docker context ls, docker ps, and docker images commands are used to inspect the host system and running services.
[COMMAND_EXECUTION]: The skill performs active validation by building and running containers based on local files, which involves executing arbitrary instructions defined within those files (e.g., RUN instructions in a Dockerfile).
Evidence in SKILL.md: docker build --no-cache -t test-build ., docker run --rm -d --name validation-test test-build, and docker exec validation-test ps aux.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests local project files that influence its execution logic and validation steps without sufficient safeguards.
Ingestion points: The skill reads project structure and file content using find . -name "Dockerfile*" and find . -name "*compose*.yml" in SKILL.md.
Boundary markers: No delimiters or specific instructions to ignore embedded commands or instructions are present when the agent processes file content.
Capability inventory: The skill has the capability to build images, run containers, and execute commands inside them via docker build, docker run, and docker exec as defined in the validation steps of SKILL.md.
Sanitization: No sanitization or validation of the ingested file content (Dockerfile, Compose files) is performed before these files are used in build or run commands.

docker-expert