# Upstream > Issues and pull requests Serge Gatezh has filed on projects he doesn't own, from Cloudflare and Bun to Playwright and Better Auth, and what came of each. When something I build on breaks, I file it where it can be fixed for everyone. This is the log. Each entry links to the issue or pull request on GitHub and says what the report was about; the line beside it says what came of it. Projects of my own are on the [projects](/projects/) page, not here. ## [umputun/agterm](https://github.com/umputun/agterm) - [Issue #915: Live sessions: quit-time capture keeps no command for any pane when the single 0.1 s zmx list times out, so a reboot restores bare shells](https://github.com/umputun/agterm/issues/915). Fixed November 6, 2026. At quit, agterm gave the session listing 0.1 s to answer. When it ran over, no pane's running command was saved, so after a reboot the panes came back as bare shells. ## [cloudflare/cloudflare-docs](https://github.com/cloudflare/cloudflare-docs) - [Issue #34799: Durable Objects migrations reference: "versions upload fails fast" with exports is outdated](https://github.com/cloudflare/cloudflare-docs/issues/34799). Open since October 30, 2026. The migrations page says `versions upload` fails fast when the config has `exports`. Current Wrangler stores them on the version and reconciles at the next deploy; the issue cites the source. ## [samuelgursky/davinci-resolve-mcp](https://github.com/samuelgursky/davinci-resolve-mcp) - [Pull request #315: fix(installer): skip Blackmagic Disk Speed Test's sandbox container](https://github.com/samuelgursky/davinci-resolve-mcp/pull/315). Merged October 23, 2026. The installer took Disk Speed Test's container for a sandboxed Resolve and wrote the bridge scripts and their auth token into it. The fix tightens the container match. ## [cloudflare/cf](https://github.com/cloudflare/cf) - [Issue #390: `cf workers …` API commands fail without workerd: workers module statically imports miniflare](https://github.com/cloudflare/cf/issues/390). Fixed November 4, 2026. `cf workers scripts list` is an API read, but the module imported miniflare at load time, so it failed wherever workerd was not installed, such as a slim devcontainer image. - [Issue #252: Support loading cloudflare.config.ts when cf runs on Bun](https://github.com/cloudflare/cf/issues/252). Declined October 20, 2026. With a `cloudflare.config.ts` in the directory, every command failed under Bun, including account-level ones that never read it. Node stays required. - [Issue #251: cloudflare.config.ts can't opt out of dependency metadata upload (dependencies_instrumentation dropped by cf migrate)](https://github.com/cloudflare/cf/issues/251). Fixed October 12, 2026. A Wrangler project could opt out of uploading its npm dependency metadata. `cf migrate` dropped that setting and the new config had nowhere to put it. ## [oven-sh/bun](https://github.com/oven-sh/bun) - [Issue #45563: install.globalStore: share entries whose peer is typed by `@types/*`, instead of keeping them project-local](https://github.com/oven-sh/bun/issues/45563). Open since October 16, 2026. After the `@types/*` peer fix, React projects stopped sharing most of their dependencies through the global store: 82 MB to 300 MB of project-local files on one monorepo. Proposes sharing those entries. ## [microsoft/playwright](https://github.com/microsoft/playwright) - [Issue #42203: captureGitInfo diff capture shallows the user's repository via `git fetch --depth=1`](https://github.com/microsoft/playwright/issues/42203). Fixed August 11, 2026. Collecting report metadata on a pull request ran `git fetch --depth=1` and turned a full clone into a shallow one. Reported with a one-test repository that fails on exactly that. ## [microsoft/playwright-mcp](https://github.com/microsoft/playwright-mcp) - [Issue #1599: `ensureTab()` returns a crashed page reference, leaving the server stuck reporting "Page crashed" until restart](https://github.com/microsoft/playwright-mcp/issues/1599). Fixed May 5, 2026. After the Chromium renderer crashed under memory pressure, the server kept handing out the dead page, so every tool call failed until the process was restarted. ## [rtk-ai/rtk](https://github.com/rtk-ai/rtk) - [Issue #1307: `rtk init` hangs in non-interactive environments (v0.36.0 telemetry consent gate)](https://github.com/rtk-ai/rtk/issues/1307). Fixed July 20, 2026. `rtk init` waited on stdin for a telemetry consent answer, which hung every Docker image build and CI job that ran it. ## [better-auth/better-auth](https://github.com/better-auth/better-auth) - [Issue #9156: `dash` plugin insight incorrectly flags custom defaultRole as non-existent](https://github.com/better-auth/better-auth/issues/9156). Fixed April 15, 2026. The infrastructure dashboard warned that a custom `defaultRole` did not exist, although it was a key of the admin plugin's own `roles` map. ## [umputun/ralphex](https://github.com/umputun/ralphex) - [Issue #72: Docker wrapper binds port even without `--serve`, preventing multiple concurrent instances](https://github.com/umputun/ralphex/issues/72). Fixed February 7, 2026. The Docker wrapper always published port 8080, so a second instance in another project failed to start even when nothing was serving. - [Issue #62: Docker wrapper mounts ~/.config/ralphex as read-only, causing startup failure](https://github.com/umputun/ralphex/issues/62). Fixed February 3, 2026. The wrapper mounted the config directory read-only, and the first run died writing its own defaults into it. Found a bug in something of mine? Open an issue on the repository, or [send me a message](/contact/).