Blog
Notes on running AI coding agents in parallel, and on keeping their rules and decisions in step.
CLAUDE.md changed mid-task: other worktrees never saw it
Does Claude Code reread CLAUDE.md mid-task? Our runs: yes in its own checkout (4 of 4), never in other worktrees (0 of 7). Numbers, a correction, how to rerun.
How trailstone fits a normal git workflow, with real cases
How trailstone fits an AI coding agents team workflow: branch, commit, PR, pre-push guard, CI. Real commands and output, six cases, and when to skip it.
Why trailstone: decisions change, agents keep building
Why trailstone exists: AI coding agents keep building on a decision you already reversed. The idea, what we measured (and got wrong), and open directions.
Run parallel Claude Code agents with git worktrees
How to run multiple Claude Code agents in parallel with git worktrees: the commands, the pitfalls, and what happens when a rule changes mid-task.
Architecture decision records for AI coding agents
Architecture decision records for AI coding agents: why docs/adr/ files go unread, what an agent needs instead, and how a superseded ADR becomes a reversal.
CLAUDE.md vs AGENTS.md: what each is and how to use both
CLAUDE.md vs AGENTS.md: which agents read which file, how to keep them in sync with an import or a symlink, and the one case a static rules file can't cover.