Overview
Systematic upgrade of the TARS Personal AI Infrastructure to leverage Claude Opus 4.6 capabilities — primarily Agent Teams (parallel Task tool), extended context compaction, and Managed Agents API.
Scope: 3 waves, 28 tasks across coordination layer, specialist tools, and fleet integrations.
Current Status
activeWave 1 (Coordination Layer) is complete. Wave 2 (Specialist Tools) in progress.
Wave Structure
Wave 1 — Coordination Layer (Complete)
- SQLite persistence for tasks, costs, and activity log
- Hook integrations (pre/post tool, notification)
- Evaluation CLAUDE.md patch
- Session state serialization
Wave 2 — Specialist Tools (In Progress)
- BlackBeltSuite Agent Teams parallelism
- Newsletter parallel research agents
- Job search campaign orchestration
- Research multi-agent synthesis
Wave 3 — Fleet Integration (Planned)
- CASE/RANGER/PLEX coordination protocols
- Cross-fleet task routing
- Shared R2 artifact passing
Key Decisions
The decision to use Postgres (Neon) for the knowledge layer is documented at decisions/2026-04-22-postgres-knowledge-layer .
Stack
- Runtime: Bun + TypeScript
- DB: Neon Postgres + SQLite (local coordination)
- Orchestration: Claude Code Task tool
- Secrets: sops + age
Links
- Cloud workspace:
~/cloud/projects/tars-upgrades/ - Task tracker:
~/cloud/projects/tars-upgrades/02-working/