AgentLoop v1.1.0

Adds workspace-write sandboxing with automatic approval review for worker and critic sessions, a zero-setup recorded-run replay, mobile replay improvements, and a reproducible independent evaluation with regression coverage. Documentation now clarifies the solo-developer audience and implementation evidence.

AgentLoop v1.0.0

First public release. A local orchestration daemon for coding agents: every cycle starts a fresh worker, work carries forward in project files, and a fresh critic enforces GUIDELINES.md with a strict PASS/FAIL contract. Ships with a one-shot dispatch API, a loop engine with optional polish mode, a live dashboard with transcripts, events, messages, and cancellation, an MCP bridge for driving runs from ChatGPT, and two example projects. Plain Node 18+, zero package dependencies, all state in local JSON/NDJSON files.