The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructions in SKILL.md require administrative access to install tools, specifically using sudo dpkg -i for the dive analysis package.\n- [EXTERNAL_DOWNLOADS]: The skill downloads binaries from external sources, including GitHub (github.com/wagoodman/dive) and Google Cloud Storage (storage.googleapis.com/container-diff). While these platforms are reputable, the binaries are executed with elevated privileges.\n- [REMOTE_CODE_EXECUTION]: The workflow includes downloading and immediate execution or installation of remote binaries (dive and container-diff), which bypasses standard package registry verification processes.\n- [COMMAND_EXECUTION]: Both SKILL.md and scripts/agent.py perform extensive shell command execution via the docker CLI and other system utilities to collect forensic evidence.\n- [DYNAMIC_EXECUTION]: The workflow in SKILL.md dynamically generates and executes Python scripts using heredocs (python3 << 'PYEOF') to parse and analyze JSON data from container artifacts.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from compromised environments, creating a potential surface for indirect prompt injection.\n
Ingestion points: Filesystem metadata from docker diff, configuration JSON from docker inspect, and container log files.\n
Boundary markers: None identified; untrusted data is processed directly by analysis scripts.\n
Capability inventory: The skill has the ability to execute shell commands and write evidence to the host filesystem.\n
Sanitization: The Python scripts use simple pattern matching but do not employ comprehensive sanitization for data that might be fed back into an LLM context.

analyzing-docker-container-forensics