git-beauti/resume-failed-compaction

★ 15PythonAudience · developerComplexity · 1/5LicenseSetup · easy

Mindmap

mindmap
  root((resume-failed-compaction))
    What it does
      Recover interrupted sessions
      Scan local Codex logs
      Generate handoff summary
    How it works
      Read JSONL session files
      Read TUI logs
      Confidence scoring
      Human-readable explanation
    Summary Contents
      Recent user requests
      Assistant replies
      Tool calls made
      Working directory
      Token pressure point
    Tech Stack
      Python
      Codex skill system

mindmap root((resume-failed-compaction)) What it does Recover interrupted sessions Scan local Codex logs Generate handoff summary How it works Read JSONL session files Read TUI logs Confidence scoring Human-readable explanation Summary Contents Recent user requests Assistant replies Tool calls made Working directory Token pressure point Tech Stack Python Codex skill system

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Recover a Codex coding session that was cut short by a failed context-compression step without losing track of what the agent was doing.

USE CASE 2

Get a structured handoff summary from an interrupted AI session to paste into a new Codex thread so work continues from the right point.

USE CASE 3

Inspect which Codex session on your machine was most likely interrupted and see a confidence score explaining why.

Tech stack

Python

Getting it running

Difficulty · easy Time to first run · 5min

Install by copying the folder into your Codex skills directory with a single command, no additional dependencies beyond Python.

Use freely for any purpose, including commercial, as long as you keep the copyright notice.

In plain English

Resume Failed Compaction is a skill for Codex, OpenAI's AI coding agent. It solves a specific problem: when an AI coding session runs long, the agent tries to compress its conversation history to stay within memory limits. If that compression process fails or drops important context, the next thread the agent starts has no idea where the previous session left off, and work gets lost. This tool addresses that by scanning your local Codex session files and logs. It looks at the JSONL files and TUI logs that Codex writes to your computer, identifies which session most likely got interrupted by a failed compaction, and generates a short summary. That summary is designed to be handed to a fresh Codex thread so it can continue the actual task rather than starting from scratch. The scanner is read-only. It does not modify session files, delete logs, or touch your repository. It assigns each candidate session a confidence score and provides a human-readable explanation of why it thinks that session failed. The summary it produces covers recent user requests, assistant replies, tool calls, the working directory, and token pressure at the point of failure. Installation means copying the folder into your Codex skills directory. On Mac or Linux that is a single cp command. On Windows there is a PowerShell equivalent. Once installed, you ask Codex to use the skill by name and it runs the scanner, finds the most likely interrupted session, and uses the recovery summary to pick up from where the work stopped. The project is MIT-licensed and the README includes both English and Chinese descriptions of what it does.

Copy-paste prompts

Prompt 1

I installed the resume-failed-compaction skill in Codex. Walk me through telling Codex to use it and getting a recovery summary after a failed compaction event.

Prompt 2

My Codex session stopped mid-task. Use resume-failed-compaction to find the interrupted session and give me a summary I can paste into a new thread to continue the work.

Prompt 3

Show me how resume-failed-compaction assigns confidence scores to candidate sessions, what signals does it look for in the JSONL and TUI log files?

Prompt 4

I want to extend this skill to also capture which files in the repository were modified before the session died. How would I add that to the recovery summary?

Open on GitHub → Explain another repo

← git-beauti on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.