Back to browse
GitHub Repository

A new coding methodology to build incredibly large things.

4 starsShell

Tarvos – fix context rot by chaining fresh Claude agents automatically

by Photon48·Mar 12, 2026·1 point·0 comments

AI Analysis

●●●BangerSolve My ProblemBig Brain

Auto-spawning fresh agents at 50% context beats manual session restarts.

Strengths
  • Git worktree isolation per session prevents merge conflicts
  • Progress handoff notes preserve context between agent swaps
  • TUI for accepting/rejecting merges without touching git directly
Weaknesses
  • Claude Code only — no support for Cursor, Continue, or other agents
  • Rough edges acknowledged by author, needs polish
Target Audience

Developers using Claude Code for multi-phase projects

Similar To

Cursor · Continue · Claude Code

Post Description

AI coding agents degrade as context fills up. By phase 6 of 8, output quality has quietly fallen apart. I kept restarting sessions by hand — copy-pasting my PRD and progress notes into a fresh context just to keep it on track. That's not an autonomous developer. That's babysitting. Tarvos automates the handoff.

When context crosses 50%, it writes a progress note and spins up a fresh agent that picks up exactly where the last one left off. Each session gets its own isolated git worktree. When the work is done, accept merges to your branch, reject discards cleanly.

You write a markdown plan. Tarvos runs it to completion without you.

Open source, MIT. Works today, rough edges exist.

Similar Projects