v1.0.0 — never-stops, always-better

A never-stopping, always-improving Ralph loop for Google Antigravity that drives the agy
headless agent CLI to improve a repo forever — where every accepted handoff is provably better
than the last.

Two guarantees

• Never stops — launchd → bash supervisor → disposable agy. Survives quota outages, crashes,
  IDE death, timeouts, and reboots. Only STOP / ralph stop / --max-iterations end it.
• Always strictly better — a git-backed multi-metric ratchet independently re-measures the
  committed tree (never trusts agent-claimed numbers), auto-reverts regressions and fabrications, and
  never lowers a floor.

Three command surfaces

• ralph CLI (init · start · once · status · stop · doctor · install-daemon · …)
• Antigravity in-IDE slash command /ralphloop (alias /ralph)
• Claude Code slash command /ralph-antigravity

Quality

Designed and hardened via multi-agent orchestration (design synthesis → parallel authoring →
6-lens adversarial review). The review caught and fixed 3 CRITICAL + 6 HIGH issues (dead verify
wiring, an agent-writable RATCHET.json command RCE, multi-minute kill-switch latency, foreign-remote
push, best-metric corruption, …). Validation: 53 tests + a bounded live smoke against real agy
(2 advances + 1 correct auto-revert observed).

MIT. See SPEC.md, docs/TECH_GROUNDING.md, docs/RUNBOOK.md.