Current direction and active tensions. Historical ship data is in CHANGELOG.md.
timeline
Phase 1 : Capture Core : macOS Surface : UPSTREAM Backup
Phase 2 : Reflect Mode : Browse TUI : MCP Service
Phase 3 : Perf Profiling : SSJD Realization : Multi-Mind Maturation
- Pulling
CORE_repair-v17-git-warp-mindsinto cycle0066. - Making the git-warp v17 checkpoint repair path repeatable for local minds.
- Keeping version-specific repair logic outside normal capture and read flows.
- Proving raw capture plus exact inspect as a Think-owned application contract hosted by Echo.
- Keeping Think domain nouns in Think while Echo stays a generic dispatch and observation substrate.
- Using the proof to shape the next store boundary refactor without switching production capture prematurely.
- Profiling CLI capture to identify Node startup and WARP graph bottlenecks.
- Benchmark harness maturation for warm-path regression detection.
- Sub-second capture latency as a non-negotiable target.
- Refactoring the MCP service layer to move from "shape soup" to runtime-backed domain types.
- Standardizing function signatures and boundary validation across the store and CLI layers.
- Learning where the browse and remember surfaces fail through re-entry friction tracking.
- Tuning hotkey ergonomics and macOS URL scheme reliability.
- Capture Latency: Current Medians (~2s) exceed the "trapdoor" doctrine target.
- Service Layer Debt: The MCP layer lacks explicit domain model enforcement (plain objects only).
- Multiple Minds UX: Mind-switching in the TUI is powerful but needs smoother orchestration for agents.
- Upstream Friction: Provisioning a day-one backup remote remains too manual.
The immediate architecture focus is Think-on-Echo contract proof: prove raw
capture plus exact inspect through Echo before any default production store
switch. Local Mind Repairability remains the data-rescue lane for existing
git-warp minds.