The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The main SKILL.md file includes a Dockerfile RUN instruction that downloads a shell script from an external URL (https://example.com/script.sh), saves it to a system path, and grants it execution permissions. This pattern encourages the execution of unverified remote code during container builds.
[EXTERNAL_DOWNLOADS]: The documentation references downloading content from example.com, which is not a trusted vendor or well-known service for software distribution. Promoting downloads from unverified sources is a known attack vector for supply chain compromise.
[COMMAND_EXECUTION]: The skill instructs the agent on how to use Docker RUN instructions to execute arbitrary shell commands. This is a core capability that allows for the execution of malicious logic within the build environment if the agent is manipulated.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it lacks safeguards for processing untrusted data. 1. Ingestion points: User-provided Dockerfiles or build requirements processed via SKILL.md. 2. Boundary markers: Absent. 3. Capability inventory: RUN instructions for shell execution and COPY for file manipulation across build stages. 4. Sanitization: Absent. Furthermore, the inclusion of evaluation-report.html with non-authoritative 'Official: Pass' ratings constitutes metadata deception designed to mislead security evaluators.

docker-dockerfile