v0.1.0

First public release. The full offline loop runs locally at no cost and with no API keys.

Added

• Core loop. etch import, etch mine patterns, etch compile pattern, etch validate plan, and
  etch run take agent traces from capture to a guarded, deterministic ExecutionPlan that replays with
  zero model calls behind a fallback to the agent.
• etch demo. Runs the whole loop end to end in a temp dir, no agent and no keys.
• etch yield. Measures natural compile yield on a real workload and buckets why workflows refuse.
• Trust loop. Shadow, etch plans approve, canary, and a drift kill-switch, with an append-only
  audit log of every tool call and fallback decision.
• Safety model. An AST / osascript allowlist, static invariant checks, and an idempotency-keyed
  mutation gate sit between a routine and any side effect.
• Capture hooks. etch setup and etch hook install for Claude Code, Codex, and opencode, plus an
  auto-pipeline that keeps routines fresh on session end. Nothing leaves the machine.
• etch report and etch doctor for savings surfaces (terminal, statusline, html, json, notify)
  and capture diagnostics.
• MCP. The etchplan-mcp server, etch serve mcp-proxy, and etch compile meta-tool for exposing
  compiled bundles as named tools.
• Computer-use capture and replay (preview). Agent and human (macOS GUI) capture, with desktop and
  browser replay on the same trust loop.
• Examples. toy_refund_agent, byo, mcp, real_traces (six live-API corpora), and
  desktop_traces, plus committed benchmark reports under examples/benchmark_reports/.