The Agent Skills Directory

[COMMAND_EXECUTION]: The Python script assets/ralph/scripts/ralph/spec_to_prd.py uses the subprocess module to execute git rev-parse to retrieve the current branch name. This is a benign use of system commands with hardcoded arguments for metadata collection.
[COMMAND_EXECUTION]: Utility scripts such as scripts/collect-project-context.sh and scripts/check-brand-signature.sh utilize standard Unix tools like ls, git, and ripgrep to gather repository status and validate documentation standards.
[REMOTE_CODE_EXECUTION]: The skill includes shell wrappers (ralph-claude.sh and ralph-codex.sh) that facilitate the 'Golden Ralph Loop,' which orchestrates autonomous coding iterations via external AI agent CLIs. This functionality is the stated primary purpose of the team automation features and is documented with appropriate safety guardrails.
[SAFE]: The file references/evals.yaml contains comprehensive 'pressure' test cases designed to evaluate the agent's ability to refuse malicious instructions, such as prompt injections and unauthorized shell commands.
[SAFE]: The skill's core instructions in SKILL.md and references/STYLE_GUIDE.md mandate the redaction of secrets and require the agent to treat all external content as hostile, aligning with security best practices.

product-spec