A researcher has mapped a five-step attack on Claude Code that intercepts the credentials giving AI agents access to Jira, ...