The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation script scripts/install.sh and documentation in SKILL.md and references/commands.md recommend installing the CLI via curl -sSL https://strix.ai/install | bash. This pattern downloads and executes arbitrary code from a remote server without verification.
[COMMAND_EXECUTION]: The skill uses wrapper scripts (scripts/run-scan.sh, scripts/ci-scan.sh) to execute the strix binary and docker commands on the host system. It requires an active Docker daemon and can pull container images from remote registries.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it is designed to ingest and analyze content from untrusted external sources like GitHub repositories and live URLs.
Ingestion points: Target URLs and repository paths are passed to the tool via the --target or -t flag in scripts/run-scan.sh and SKILL.md.
Boundary markers: The skill suggests using --instruction-file for scoping and rules of engagement, but this does not provide a technical boundary to prevent instructions embedded in target data from influencing the agent's behavior.
Capability inventory: The skill is granted Bash and WebFetch tools and interacts with the system's Docker environment.
Sanitization: There is no evidence in the skill's wrapper scripts that content from external targets is sanitized or filtered before being processed by the underlying AI-driven tool.

strix