Brief, mostly facts

bunker: sandboxed Claude agents wrote 159 of 188 PRs

An MIT-licensed label-driven loop runs Claude Code in Docker worktrees, with a second-session review and auto-merge via CI. It costs about $1.30 per issue on Sonnet.

Read the source at reddit.com

What the research found

Developer built bunker, an open-source (MIT) label-driven agent loop that runs Claude Code in Docker worktrees, reviews changes with a second session, and auto-merges via required CI; about 160 of its first 190 PRs were self-written at about $1.30 per issue on Sonnet, now v0.11 and Claude-only on Linux/macOS.

Claims and how well the evidence supports them

Bunker wrote about 160 of its first 190 PRs 95%
An issue costs around $1.30 of Sonnet time on average 99%
Bunker is at v0.11, Claude only, Linux/macOS with Docker, MIT licensed 99%
Fine-grained GitHub tokens cannot read check runs 97%

Evidence

reddit.com https://reddit.com/r/ClaudeAI/comments/1xagj87/sandboxed_claude_agents_wrote_159_of_my_188_prs
github.com https://github.com/dbggr/bunker
bunker.true.engineer https://bunker.true.engineer

Actionability rated low.