matthew-gerstman

PR #7045 Review: Add Back Dokploy Preview Environments

Title: Feat/add back dokploy preview envs Author: Elroy Fernandes Size: +1397 / -453 across 11 files, 55 commits

Summary

This PR restores Dokploy-based preview environments for PRs. Each PR gets an isolated stack (API, dashboard, Postgres, Redis, LocalStack, Inngest) deployed to a worker server, with Traefik routing from a manager node. It includes dynamic port allocation for multi-PR coexistence, a stale preview cleanup cron, and a rewritten lambda deployer that uses the real deploy-lambda.sh scripts.

matthew-gerstman

Agent Streaming Performance Research

Problem Statement

Agent responses in the Dashboard UI arrive one word at a time with a long delay before the first word. This is a regression over the past few weeks, exclusively in local dev — production/staging are fine.

---

Architecture Overview

matthew-gerstman

Project Folders Cleanup Plan

Context

The project folders feature (workspace-scoped folders that organize projects) shipped as an MVP and has accumulated technical debt. The three main areas needing attention are: incomplete error handling in dashboard components, service-layer code quality issues, and gaps in test coverage. This plan covers a full cleanup pass.

---

Commits

matthew-gerstman

Upgrade CLI to React 19 + Ink 6

Context

The CLI is isolated from the monorepo (excluded from workspaces in root package.json, has its own cli/bun.lock) solely because Ink 5 required React 18 while the dashboard uses React 19. Ink 6 now supports React 19, so the version conflict no longer exists. This upgrade unblocks agents that trip over the React 18 isolation and simplifies the monorepo setup.

Commits

1. feat(cli): upgrade to React 19 + Ink 6
2. refactor(cli): integrate CLI into monorepo workspaces

matthew-gerstman

SSE Resilience Improvements

Context

The SSE implementation is already well-architected with exponential backoff, heartbeat monitoring, event replay via lastEventId, and Redis-backed connection tracking. However, there are gaps in failure recovery paths — specifically around unbounded replay queries, silent replay failures, a heartbeat race condition, and lack of Redis subscriber health monitoring. These gaps mean that after prolonged disconnections or infrastructure hiccups, clients can end up with stale data and no way to detect or recover from it.

Improvements (ranked by impact-to-risk ratio)

1. Bound the event replay query + notify client of truncation

Problem: getRecentEventsFromDatabase() has no LIMIT clause. A reconnecting client with a stale/missing lastEventId can trigger an unbounded query returning all events for a topic. One reconnecting client can spike database load for everyone.

matthew-gerstman

# SSE Resilience Improvements — Future PRs

Findings from investigating dropped agent events in PR #6692.

## 1. Event replay gap for thread status updates (HIGH)

`emitAgentStatusEvent` in `apps/api/src/agents/obvious-v2/state/events.ts` calls
`publishToProjectUsers` without a `tx` parameter, so `thread:updated` events aren't
stored for replay. If a client reconnects during an agent run, they miss status
transitions and the UI gets stuck on "thinking" or "running".

matthew-gerstman

Abstract Messaging: Slack → Multi-Platform (SMS/Twilio first) with Group Chat

Commits

1. feat: add messaging platform types, adapter interface, and registry
2. feat: add database tables for generic messaging connections and links
3. feat: add thread/message service methods for generic messaging links
4. refactor: create Slack messaging platform adapter wrapping existing code
5. feat: add Twilio SMS/Conversations service wrapper and adapter
6. feat: add SMS inbound webhook route and connection setup
7. feat: add messaging event dispatcher with group chat support

matthew-gerstman

# Follow-up: Claude Bot Review Comments on PR #6641

## Context
PR #6641 adds employee mode credit bypass. Claude bot flagged three issues.
Two were addressed in the PR; the "critical" one is a false positive.

## 🔴 "Critical": Per-step credit deduction not bypassed
**Status: False positive — no action needed**

Claude claims the per-step deduction block still runs for employee mode.

matthew-gerstman

Obvious Monorepo — Comprehensive Test Coverage Plan

Date: 2026-02-08 Current coverage: ~25-30% of source files have corresponding tests Coverage enforcement: None (no thresholds, no CI reporting)

---

Executive Summary

matthew-gerstman

SSE Event Stream Performance Analysis

Date: February 8, 2026 Runtime: Bun 1.3.9 (darwin arm64) Method: Custom stress test suite using Bun's built-in --cpu-prof-md and --heap-prof-md profiling tools

---

Executive Summary