How to use Codex to refine coding guidance
Codex extracted and analyzed pi sessions to shape an outer guidance prompt; the author reported mixed results and reduced check-in time on a game.
news.ycombinator.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Extract my prompts from [session archive], separating them from agent-to-subagent messages. Analyze recurring instructions and draft a flowchart for an outer guidance prompt that prioritizes hard work, ongoing refactoring and reflection.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- athrowaway3z
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The author reported game-development check-in time falling to one fifth of its previous level, while warning that results were mixed and reduced involvement might miss good ideas.
- Human involvement
- The author continued checking in on the game, reported mixed results, and withheld trust for delicate codebases.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Had Codex write a pi session extraction tool.
- Filtered personal prompts from agent and subagent conversation.
- Had Codex analyze prompting patterns and turn them into a guidance flowchart.
The author used Codex to examine their own coding-agent interactions and turn recurring behavior into guidance for further work. First, they had Codex write a tool that extracted their pi sessions. That extraction needed filtering to separate the author's prompts from agents talking to subagents. They then asked Codex to analyze patterns in those prompts and turn the findings into a flowchart for an outer guidance-creating prompt.
The motivation was a workflow that had become mechanical: repeatedly telling the agent to do something and then review it. The author describes reminders to tackle hard work first, clean up and refactor along the way, and reflect after the first return. That reflection feeds information about unfinished work back into the guidance prompt, which assigns more work.
The reported result was mixed. For a game under development, check-in time fell to one fifth of the previous level. The author did not trust this approach with delicate codebases and worried that spending less time involved could mean missing good ideas. Who it's for: developers interested in using their own session history to improve recurring coding instructions.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · athrowaway3z
Frequently asked questions
What session data did Codex analyze?
The author had Codex extract pi sessions and filter their prompts from conversations between agents and subagents.
Did the author trust the resulting workflow?
Only to a limited extent: results were mixed, and the author said they did not trust it with delicate codebases.
More details
About this account
Codex wrote the extraction tool and analyzed interaction patterns to shape guidance for subsequent agent work.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=48644489. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: athrowaway3z on Hacker News
Source published
Added to Agents In Use
Source checked
Tools and topics
Codex · pi
Agent workflows
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.