The CI and test suite have recurring performance issues that block development velocity and make contribution harder.

Test worker memory

• `prepare-pr` uses `unit-fast` config with long-lived workers that accumulate 4GB+ per process over their lifetime. Kills 16GB machines. GC pressure causes crawling or crashes.
• Workaround: splitting into 2x chunks based on time estimate (dropped from 10s to 5s per chunk estimate). Helped substantially but not a root fix.

CI reliability

• CI regularly red for multiple days. Multiple maintainers spend hours daily shepherding it green.
• Large refactors frequently break CI — acknowledged pattern.
• Barrel imports in extension tests were causing timeout failures (recently fixed).
• `pnpm check` had become a "giant kitchen sink" (recently cleaned up).
• `checks-fast-extensions` was chronically red (recently stabilized).

Developer experience

• Running the full test suite locally requires high-end hardware (Mac Studio class).
• No guidance on which test configs to use for different development scenarios.
• Missing AGENTS.md in key directories means coding agents (Codex, Claude Code) cross boundaries and create spaghetti — boundary docs proposed in Docs: add boundary AGENTS guides #56647.

What's working

• Release automation (server-triggered releases) reduces human-in-loop failure modes.
• Multiple maintainers actively fixing CI.
• Test sharding improvements reducing per-chunk time.

Suggested workstreams

• Test worker memory: profiling and lifecycle optimization for long-lived workers
• AGENTS.md boundary docs across key directories (Docs: add boundary AGENTS guides #56647)
• CI reliability dashboard or automated green-tracking
• Developer experience docs: which configs for which scenarios, minimum hardware requirements
• Extension test isolation: prevent barrel imports and cross-boundary dependencies from causing timeouts

This is a tracking issue — feel free to pick up any workstream or propose new ones.