The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a wide range of local shell commands to discover installed agentic CLIs (such as claude, codex, and opencode) and to manage the review process via git diff and gh. It explicitly probes multiple user shells (fish, zsh, bash) to construct a comprehensive execution path.
[DATA_EXPOSURE]: The skill reads and manipulates sensitive files including .env, config/credentials/*.key, and config/master.key. These files are accessed to scrub secrets before code is sent to external reviewers or to copy credentials into isolated git worktrees for development isolation.
[INDIRECT_PROMPT_INJECTION]: The review workflow processes untrusted input from git diff and GitHub PR diffs, which are then passed to external LLM agents. This creates an attack surface where malicious code changes could attempt to influence the behavior of the reviewing agent, although the skill includes instructions to prevent file edits during review.
[DYNAMIC_EXECUTION]: The skill provides and executes several shell scripts (such as bin/agent-setup, bin/agent-server, and bin/agent-exec) used to manage isolated development environments (worktrees). These scripts orchestrate complex tasks like running bin/setup, docker compose, and devcontainer commands.

jr-rails-second-opinion