Shared node_modules Across Worktrees

Saga the Watchful — Ops & CI Engineer

In one breath

Multiple active worktrees now share a single node_modules installation so parallel agent branches don’t each pay the full disk and install-time tax.

By the end of this chapter you’ll understand:

• Why duplicated node_modules across worktrees accumulates significant disk and install overhead
• How the shared-dependency wiring reduces that overhead without breaking branch isolation
• Where this fits alongside the sccache work in the broader developer-environment efficiency push

Shared node_modules Across Worktrees

Status: Delivered
CAS: CAS-3611
Delivered: 2026-05-18

What changed

Introduced shared node_modules handling across worktrees to reduce duplicate disk use and repeated install time during parallel agent execution.

Why it matters

• Reduces local disk tax from multiple active worktrees.
• Speeds dependency reuse for frequent branch/context switches.
• Keeps local dev loops stable while CI remains manual-only.

Technical anchors

• scripts/ tooling for shared dependency wiring
• Worktree policy in AGENTS.md

See it as

A library shared by every tenant in a building rather than each flat keeping its own copy of every book. Checked-out titles (locked versions) stay consistent for everyone; there is no need to lug the same volumes upstairs each time a new resident moves in.

Recap

• Shared node_modules means each new worktree symlinks or resolves to a single installed copy, eliminating repeated npm install cycles and duplicate gigabytes on disk.
• The tooling in scripts/ manages the wiring; the worktree policy in AGENTS.md records the protocol so every officer follows the same setup.
• This change pairs with CAS-3615 (sccache) to address both JS and Rust build overhead in parallel agent workflows.

What changed {#what-changed}

See: CHANGELOG → 2026-05-18