Verify, in a FRESH environment (Docker container or throwaway $HOME), that a brand-new user running rig init gets a fully-working setup with ZERO manual steps:
- skills are discoverable by the harness —
~/.claude/skills/<skill> symlinks resolve AND appear in the Claude Code Skill-tool list (not just sitting in ~/.agents/skills);
- agent-hooks + global git-hook dispatcher installed and firing;
- CI gate workflows + harness auto-mode (.claude/settings.json) provisioned;
- idempotent re-apply is a no-op;
rig status clean.
Today's rig tests use a tmp-$HOME unit sandbox (good) but there's no full clean-room integration test that proves 'any new user on any machine' works. This is the gold-standard verification the skill-loading bug (rig-cli#9) showed we need — a manual symlink fixed THIS machine; only a clean-room e2e proves it's universal.
Verify, in a FRESH environment (Docker container or throwaway $HOME), that a brand-new user running
rig initgets a fully-working setup with ZERO manual steps:~/.claude/skills/<skill>symlinks resolve AND appear in the Claude Code Skill-tool list (not just sitting in ~/.agents/skills);rig statusclean.Today's rig tests use a tmp-$HOME unit sandbox (good) but there's no full clean-room integration test that proves 'any new user on any machine' works. This is the gold-standard verification the skill-loading bug (rig-cli#9) showed we need — a manual symlink fixed THIS machine; only a clean-room e2e proves it's universal.