The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: In references/mutation-frameworks.md, the installation procedure for the Mull framework specifies piping a remote shell script directly into a shell interpreter (curl -1sLf '...' | sudo -E bash), which allows untrusted code execution if the endpoint or communication channel is compromised.
[PRIVILEGE_ESCALATION]: The skill commands the agent to use sudo within references/mutation-frameworks.md to create system directories, copy binaries, and execute piped installation scripts, exposing the environment to potential privilege escalation risks.
[EXTERNAL_DOWNLOADS]: The documentation instructs downloading installer scripts and prebuilt binaries from external locations such as dl.cloudsmith.io and third-party GitHub repositories for tool setup.
[COMMAND_EXECUTION]: The skill heavily relies on running external commands and installing package toolchains locally (via uv tool install, cargo install, pnpm add, and go install) to execute various mutation testing utilities.
[INDIRECT_PROMPT_INJECTION]: The skill implements an indirect prompt injection attack surface during Phase 3 triage, where it ingests external codebase data and test report outputs (such as mutmut, Stryker, and necessist outputs) into the agent's context without defining strict delimiters or validation constraints.

genotoxic