The Agent Skills Directory

EXTERNAL_DOWNLOADS (HIGH): The skill downloads binaries from github.com/peteonrails and clones repositories from aur.archlinux.org. These sources are not in the trusted whitelist, making the integrity of the downloaded content unverifiable.
REMOTE_CODE_EXECUTION (HIGH): The skill uses curl to download a binary, grants execution permissions with chmod +x, and runs it directly. This pattern of downloading and executing untrusted code at runtime is a high-risk security vulnerability.
COMMAND_EXECUTION (MEDIUM): The skill executes local shell scripts (./scripts/build-docker.sh) and uses Docker to mount host directories (-v $(pwd)/releases:/releases). This allows the container and any code running within it to interact with the host filesystem.
PROMPT_INJECTION (HIGH): The skill has a high attack surface for indirect prompt injection (Category 8). It ingests untrusted data from external sources and processes it using high-privilege tools like Bash and docker. Evidence:
Ingestion points: github.com release URLs and aur.archlinux.org git repositories.
Boundary markers: Absent.
Capability inventory: Bash tool (curl, chmod, execution), docker run (privileged directory mounting).
Sanitization: Absent.

docker-test