v0.1.6

What's Changed

• fix(machine): Resources panic paths -> Option/Result, dispatch push hitch, encode Hitch by @yzxoi in #60
• fix(machine): encode System/Assistant Hitch so LLM sees past errors by @yzxoi in #61
• feat(server): gRPC server with Open/Step/Destroy RPCs by @EricSanchezok in #58
• fix(ci): install protoc for prost-build compilation by @yzxoi in #69
• feat(proto,server): populate content_text, tool/model info, add cross-layer consistency tests by @yzxoi in #71
• refactor(machine,accelerator): remove Phase abstraction, inline setup into Captain state machine by @yzxoi in #75
• feat(machine): multi-modal encode/decode support with convenience constructors by @yzxoi in #76
• feat(proto,server,mcp): multi-modal proto transport, decode, and MCP data URLs by @yzxoi in #77
• fix(accelerator): Captain first_call flag pollution by env_action by @yzxoi in #79
• remove Accelerator.purpose_override — use State dataflow instead by @EricSanchezok in #78
• fix(machine): Context Result-ify insert/replace/remove/swap; dispatch pushes hitch by @yzxoi in #80
• fix(accelerator): cap shell output reads at 512 KB by @yzxoi in #81
• fix(machine): parallel tool execution in reactor via join_all by @yzxoi in #82
• fix(machine): isolate tool panics via tokio::spawn in reactor by @yzxoi in #84
• fix(python-sdk): decouple destroy() from channel close by @yzxoi in #85
• fix(machine): use first real message as initial prompt, not literal . by @yzxoi in #86
• fix(machine): restore .messages() for context beyond first fragment by @yzxoi in #88

Full Changelog: v0.1.3...v0.1.6