The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill references and encourages fetching configuration files and templates from several raw.githubusercontent.com URLs under the canonical organization. Since canonical is not in the explicitly trusted GitHub Organizations list, these are classified as unverifiable external downloads.
COMMAND_EXECUTION (MEDIUM): The skill executes scripts/create-charmkeeper-vm.sh, which performs significant system-level operations including multipass launch, snap install, and juju bootstrap. These commands can significantly alter the host or VM environment.
PRIVILEGE_ESCALATION (MEDIUM): The script scripts/create-charmkeeper-vm.sh uses sudo within the virtual machine to install snaps with classic confinement and prepare the k8s environment. Although restricted to the VM, it represents high-privilege operation patterns.
INDIRECT_PROMPT_INJECTION (LOW): The skill is designed to ingest and process code from a repository (integration tests) and external templates. It possesses command execution capabilities (tox, multipass exec), which creates an attack surface for indirect instructions in the processed data.
Ingestion points: External template URLs listed in SKILL.md and repository test files.
Boundary markers: None (instructions do not specify delimiters for untrusted code blocks).
Capability inventory: Subprocess calls for multipass, snap, juju, and tox in scripts/create-charmkeeper-vm.sh and SKILL.md.
Sanitization: No sanitization or validation of repository content or external templates before execution.

charmkeeperration-tests