Agent Context Patch: reuse verified Codex mistakes across fresh tasks

[Cherwayway]

A fresh Codex task can repeat a mistake that an earlier task already fixed and verified.

I built Agent Context Patch to turn that exact moment into a small, reviewable repository rule:

verified failure -> Agent-owned lesson -> exact workspace patch -> fresh-task reuse

It works with OpenAI Codex and Claude Code, stays local, and stores no raw chats. The Agent owns semantic judgment; a narrow deterministic kernel owns paths, hashes, conflicts, approval boundaries, and rollback.

74-second fail / repair / reuse demo:
https://github.com/Cherwayway/agent-context-patch/blob/main/docs/launch/terminal-demo.md

Current immutable release (v0.5.4):
https://github.com/Cherwayway/agent-context-patch/releases/tag/v0.5.4

The project is MIT-licensed and unofficial; it is not affiliated with or endorsed by OpenAI or Anthropic.

I am looking for feedback from people who have actually seen a coding Agent repeat a verified mistake: where does this workflow feel too heavy, and what would stop you before first value?