1,930 posts

IO

@IliyaOblakov

Rails engineer. I ship with Claude Code and Codex: one writes, one reviews. Real workflow, real diffs, real costs. Two Engines Kit below.

Citizen of the WORLD

Pinned
IO
@IliyaOblakov
Oct 21
5 questions about my Claude Code and Codex setup. Answers from what I've been building since the first Codex and Claude CLIs: 1) Why Claude Code writes 2) What if they disagree 3) What Codex catches 4) Does the reviewer patch code 5) Who writes the tests
7
IO
@IliyaOblakov
5h
The two sessions that review my agents' memory can't write to it. One proposes, a fresh one checks, a script writes what passed. Anthropic's checklist wants the same split: reads and writes in separate tools. I review MCP connectors against it.
MCP Connector Review Against Anthropic's Checklist
From rayastats.com
IO
@IliyaOblakov
6h
Four test-only commits hit my merge queue unreviewed. One was already merged. The verdict is a file named after the exact commit. Two Engines Kit ships that Claude Code and Codex script, built on what I've gathered since the first Codex and Claude CLIs.
Two Engines Kit by RayaStats
From rayastats.com
IO
@IliyaOblakov
10h
Hey @grok, am I the only one who hasn't received a Grok @bot reset after Elon's promise of better efficiency?
1
IO
@IliyaOblakov
Nov 8
In my setup, Claude Code and Codex share a memory that can't forget. The first time a session proposed fixes to it, one in nine was wrong. The memory is append only. Notes never get edited or deleted. Facts change, so corrections get appended. A new note marked SUPERSEDES
5

IO’s posts