Gemini CLI nightly releases failed on a 0.1MB difference: 70MB of stale esbuild chunks

Likhith-BlueLotus · ghdev · 2026-11-09

A P0 PR against google-gemini/gemini-cli fixes nightly releases that had failed every night since 2026-11-05 with npm error 413 (Request Entity Too Large) when publishing.

Root cause: esbuild emits content-hashed chunks into bundle/ without ever clearing it, and the release job bundles twice (once via prepare, again after the version bump). About 70MB of the 106MB of shipped JavaScript was unreachable output from earlier builds — 4 full copies of the chunk graph. The tarball crept from 25.1MB to 25.2MB, which base64-encodes just over App Engine's 32MiB request cap.

The fix is a single rmSync before builds, halving the tarball. A textbook case of hashed build directories silently accumulating garbage until hitting an unrelated hard limit.

Original post →

More from coding & agent

coding & agent channel →