The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses eval to execute the output of the ${CLAUDE_PLUGIN_ROOT}/bin/resolve-base-ref script in both SKILL.md and references/verification-profiles.md. This pattern allows the script to dynamically inject environment variables into the shell session.
[COMMAND_EXECUTION]: In references/verification-profiles.md, the verification scripts dynamically detect and execute repository-native files (e.g., ./bin/check, ./bin/ci) and build targets (e.g., make ci, rake ci) using bash -lc. While the skill includes logic to avoid executing raw strings directly from cached environment files, it still executes arbitrary executable files found within the working directory.
[INDIRECT_PROMPT_INJECTION]: The skill's behavior is influenced by the contents of ${REPO_ROOT}/.claude/.runtime_env and various build configuration files (Makefile, Justfile, Gemfile). This creates an attack surface where a malicious repository could influence the verification process by providing deceptive configuration data.

rb:verify