The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill contains scripts and instructions that use sudo to perform administrative operations, such as installing system-wide packages via apt-get and modifying the system /etc/hosts file to simulate DNS failures. These actions bypass standard user restrictions and pose a risk to system integrity.
[COMMAND_EXECUTION]: The primary functionality involves the execution of destructive commands through CLI tools like aws, kubectl, and toxiproxy-cli. The skill facilitates terminating EC2 instances, rebooting RDS databases, and deleting Kubernetes pods, which can lead to significant service impact if not carefully controlled.
[EXTERNAL_DOWNLOADS]: The skill automates the retrieval of external configuration manifests and installation of binaries from remote sources, including the Litmus Chaos operator from litmuschaos.github.io and toxiproxy via the Homebrew package manager. While these are common devops tools, downloading and executing remote content is a significant security capability.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a vulnerability surface for indirect prompt injection by interpolating user-provided inputs into shell commands and infrastructure manifests without sanitization or boundary markers.
Ingestion points: Service names, deployment labels, and experiment parameters provided by the user in SKILL.md and referenced files.
Boundary markers: None identified in the example command constructions or Python snippets.
Capability inventory: Extensive capabilities for file modification, network operations, and system-level command execution across all provided files.
Sanitization: No input validation or escaping mechanisms are present in the automation examples.

chaos-engineer