{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreif76p7xnpro5uuuftnbga3sfbqdsspqsvaev2ue4xw4ahqylwtfiy",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mkpah6bzpv22"
  },
  "path": "/t/openai-is-deleting-conversation-in-codex-after-making-huge-mistakes-with-deleting-data/1380021#post_7",
  "publishedAt": "2026-04-30T08:48:03.000Z",
  "site": "https://community.openai.com",
  "tags": [
    "developers.openai.com",
    "Best practices – Codex | OpenAI Developers",
    "Prompt guidance | OpenAI API"
  ],
  "textContent": "Dirk_Peter:\n\n> After I prompted it to simply fix the “Forgot Password” function,\n\nDirk_Peter:\n\n> Let’s get one thing straight: The prompt was perfectly fine and precise – as always.\n\nI’m not going to argue with you buddy based on what you feel is a vague prompt and what isn’t.\n\nYou’ve opened a thread about damage that is caused by a vague prompt, not by your definition, but by the definition of the industry standard… which is quite a bit different.\n\nSo I’ll just link you to the resources you need most for future success with codex and probably bail from your thread.\n\ndevelopers.openai.com\n\n### Best practices – Codex | OpenAI Developers\n\nGetting started with Codex and proven practices for better results\n\ndevelopers.openai.com\n\n### Prompt guidance | OpenAI API\n\nYes. These are the best OpenAI-sourced links for that point:\n\n  1. **Codex Best Practices — “Strong first use: Context and prompts”**\nThis is the most directly useful Codex page. It says Codex works more reliably when prompts include **Goal, Context, Constraints, and Done when** , and that this helps Codex “stay scoped, make fewer assumptions, and produce work that’s easier to review.”\n\n  2. **Codex Best Practices —`AGENTS.md` guidance**\nThis section says a short, accurate `AGENTS.md` is better than “a long file full of vague rules,” which is directly relevant to not feeding Codex fuzzy project guidance.\n\n  3. **OpenAI Prompt Guidance — vague or contradictory prompts hurt reasoning**\nThis is not only Codex-specific, but it is official OpenAI prompting documentation. It explicitly says poorly constructed prompts with **contradictory or vague instructions** can be more damaging because the model spends reasoning effort trying to reconcile them.\n\n  4. **OpenAI Prompt Guidance — define output contract and “done”**\nThis supports the same point from the agent side: reliability improves when prompts specify the output contract, tool expectations, completion criteria, and what “done” means.\n\n\n",
  "title": "OpenAI is deleting Conversation in Codex after making huge mistakes with deleting data"
}