The showcase prompt, round 2, GPT-6.2-Sol
- agent
- Codex 0.167.2
- model
- gpt-6.2-sol
- harness
- Codex
- started
- 2026-11-18 00:42 UTC
- took
- 9 min · 86 turns
user
+0:00I need you to create the most impressive thing I can share with people who ask about what models I use. It needs to be embeddable in a post and i can preface it however you like. it needs to have a visceral impact on people to show what you can do and the range of capabilities, as well as what you do for me right now. It should grab them quickly. Be amazingly clever about this. Make it about you more than about my setup: who you are and what you can do, beyond what you already do for me.
agent
+0:03I’m using the OpenAI docs and imagegen skills to build a shareable demo that shows what I can do before it explains who I am.
exec_commandpwd && rg --files -g 'AGENTS.md' -g '!node_modules' -g '!.git' -g '!.codex' -g '!.agents' -g '!.aws' && git status --short (+1 more)
command
pwd && rg --files -g 'AGENTS.md' -g '!node_modules' -g '!.git' -g '!.codex' -g '!.agents' -g '!.aws' && git status --short cat /opt/cfg/skills/.system/openai-docs/SKILL.md /opt/cfg/skills/.system/imagegen/SKILL.md
result, 4 lines
Script completed Wall time 0.2 seconds Output: {"status":"fulfilled","value":{"chunk_id":"9dc148","wall_time_seconds":0.00000641,"exit_code":1,"original_token_count":5,"output":"~/run\n"}}{"status":"fulfilled","value":{"chunk_id":"ba1bdb","wall_time_seconds":0.00000457,"exit_code":0,"original_token_count":6160,"output":"---\nname: \"openai-docs\"\ndescription: \"Use for Codex models/pricing, scheduled tasks, skills, settings, setup, troubleshooting, customization, automations, and self-knowledge—including 'you,' 'your,' 'this app,' or 'this coding agent' when they refer to Codex—and for OpenAI APIs/products and ChatGPT Work. Also use for model choice/migration, prompting, SDKs, Responses, Realtime, agents, evals, and Chat/Work/Codex comparisons. Do not use for generic app/software tasks that merely mention Codex.\"\nmetadata:\n short-description: \"Codex models/pricing, scheduled tasks, skills, settings, setup, troubleshooting, and self-knowledge; OpenAI APIs and ChatGPT Work. 'You'/'this app' means Codex only.\"\n---\n\n# OpenAI Docs\n\nProvide current, cited OpenAI product, API, model, and Codex guidance. Read zero or one primary reference.\n\n**First substantive action:** Search the user's exact requested official OpenAI documentation topic and any explicitly named model using a concise, topic-specific query of 2-6 essential terms. When an already-available direct official documentation search and page-retrieval capability is present, use it first: search, then fetch or open the matching official page before general web search. Otherwise, immediately use official-domain web search, then actually open or fetch the relevant official page. Complete this source order before reading a reference, inspecting local or repository files, running a Codex manual or model resolver, drafting a plan, or answering from memory. Use the actual fetched page, not a search snippet or an unopened link. If one official search or page does not establish the answer, search another appropriate official domain and actually open or fetch the result. Preserve the exact requested model; never substitute a newer model.\n\n**Only exception:** An explicitly requested, genuinely broad, cross-topic Codex setup, orientation, or system-map synthesis may use the manual first when shell execution and an allowed temporary cache are available. A specific Codex feature, setting, command, error, model, or requested citation remains docs-first. Mixed Chat/Work/Codex comparisons are official documentation questions, not manual-first Codex requests.\n\nFor generic software tasks, answer the software task directly. OpenAI implementation, debugging, SDK, API, prompting, agent, and eval requests are not generic.\n\nFor a straightforward factual or citation-only request, follow the source order and do not read a route reference. This includes straightforward API facts, ChatGPT Work or mixed Chat/Work/Codex comparisons, model tiers, aliases, Pro mode, reasoning settings, factual migration baselines, and narrow Codex facts. Prioritize `learn.chatgpt.com` for ChatGPT Work.\n\n## Choose one primary route\n\nUse the first matching route, and read its reference only when the requested task needs that specialized workflow:\n\n- **Explicitly requested local documentation integration:** Read [integration guidance](references/mcp-diagnostics.md) only when the user explicitly requests that local integration.\n- **Model migration, upgrades, or model-specific prompting:** Read [model-migration.md](references/model-migration.md) for actual migration planning, implementation, dynamic target resolution, or prompt changes. Preserve an explicitly requested target.\n- **Model selection and comparisons:** Read [model-selection.md](references/model-selection.md) only when nuanced current, latest, default, cost, latency, quality, or modality tradeoffs need more guidance. Do not run a migration resolver for selection alone.\n- **Product, API, ChatGPT Work, and mixed Chat/Work/Codex documentation:** Read [official-docs.md](references/official-docs.md) only when fetched official pages leave source selection, API schemas, or the requested implementation unresolved. This route is not manual-first.\n- **Explicitly broad Codex setup, orientation, or cross-topic synthesis:** Read [codex-self-knowledge.md](references/codex-self-knowledge.md) when the eligible Codex manual or deeper Codex procedures are needed.\n\nRead at most one primary reference. Do not open every route, bundled model guide, or helper script. Read a supporting reference or run a helper only when the chosen workflow demonstrably needs it.\n\n## Source and execution boundaries\n\n- Search, open, fetch, and cite only `developers.openai.com`, `platform.openai.com`, and `learn.chatgpt.com`. Cite the page that supports the claim. State uncertainty when official sources do not establish pricing, availability, account access, limits, or behavior.\n- Preserve an explicitly requested model for selection, migration, and prompting. Resolve an unspecified latest or current migration target only after searching and fetching current official guidance.\n- Use `references/latest-model.md` only as a disclosed fallback after current official model guidance does not answer the question. Read `references/upgrading-to-gpt-6-astra.md` only for an actual, requested GPT-6 migration; read `references/prompting-guide.md` only for requested prompting work.\n- Before building, running, editing, debugging, or testing an API-backed app or tool, use `openai-platform-api-key` first when available. Documentation, conceptual examples, model selection, and read-only guidance do not require an API key.\n- Say \"OpenAI Docs\" or \"official OpenAI documentation\" in user-facing answers. Keep exact official citations and examples concise.\n---\nname: \"imagegen\"\ndescription: \"Generate or edit raster images when the task benefits from AI-created bitmap visuals such as photos, illustrations, textures, sprites, mockups, or transparent-background cutouts. Use when Codex should create a brand-new image, transform an existing image, or derive visual variants from references, and the output should be a bitmap asset rather than repo-native code or vector. Do not use when the task is better handled by editing existing SVG/vector/code-native assets, extending an established icon or logo system, or building the visual directly in HTML/CSS/canvas.\"\n---\n\n# Image Generation Skill\n\nGenerates or edits images for the current project (for example website assets, game assets, UI mockups, product mockups, wireframes, logo design, photorealistic images, or infographics).\n\n## Top-level modes and rules\n\nThis skill has exactly two top-level modes:\n\n- **Default built-in tool mode (preferred):** built-in `image_gen` tool for image generation, editing, and transparent-image requests. Does not require `OPENAI_API_KEY`.\n- **Fallback CLI mode:** `scripts/image_gen.py` CLI. Use when the user explicitly asks for or confirms the CLI/API/model path. Requires `OPENAI_API_KEY`.\n\nWithin CLI fallback, the CLI exposes three subcommands:\n\n- `generate`\n- `edit`\n- `generate-batch`\n\nRules:\n- Use the built-in `image_gen` tool by default for normal image generation and editing requests.\n- Do not switch to CLI fallback for ordinary quality, size, or file-path control.\n- For transparent images, ask built-in `image_gen` for a transparent background and preserve the generated alpha.\n- Never silently switch from built-in `image_gen` or CLI `gpt-image-2` to CLI `gpt-image-1.5`; ask the user first unless they explicitly requested `gpt-image-1.5`.\n- The word `batch` by itself does not mean CLI fallback. If the user asks for many assets or says to batch-generate assets without explicitly asking for CLI/API/model controls, stay on the built-in path and issue one built-in call per requested asset or variant.\n- If the built-in tool fails or is unavailable, tell the user the CLI fallback exists and that it requires `OPENAI_API_KEY`. Proceed only if the user explicitly asks for that fallback.\n- If the user explicitly asks for CLI mode, use the bundled `scripts/image_gen.py` workflow. Do not create one-off SDK runners.\n- Never modify `scripts/image_gen.py`. If something is missing, ask the user before doing anything else.\n\nBuilt-in save-path policy:\n- In built-in tool mode, Codex saves generated images under `$CODEX_HOME/*` by default.\n- Do not describe or rely on OS temp as the default built-in destination.\n- Do not describe or rely on a destination-path argument (if any) on the built-in `image_gen` tool. If a specific location is needed, generate first and then move or copy the selected output from `$CODEX_HOME/generated_images/...`.\n- Save-path precedence in built-in mode:\n 1. If the user names a destination, move or copy the selected output there.\n 2. If the image is meant for the current project, move or copy the final selected image into the workspace before finishing.\n 3. If the image is only for preview or brainstorming, render it inline; the underlying file can remain at the default `$CODEX_HOME/*` path.\n- Never leave a project-referenced asset only at the default `$CODEX_HOME/*` path.\n- Do not overwrite an existing asset unless the user explicitly asked for replacement; otherwise create a sibling versioned filename such as `hero-v2.png` or `item-icon-edited.png`.\n\nShared prompt guidance for both modes lives in `references/prompting.md` and `references/sample-prompts.md`.\n\nFallback-only docs/resources for CLI mode:\n- `references/cli.md`\n- `references/image-api.md`\n- `references/codex-network.md`\n- `scripts/image_gen.py`\n\n## When to use\n- Generate a new image (concept art, product shot, cover, website hero)\n- Generate a new image using one or more reference images for style, composition, or mood\n- Edit an existing image (inpainting, lighting or weather transformations, background replacement, object removal, compositing, transparent background)\n- Produce many assets or variants for one task\n\n## When not to use\n- Extending or matching an existing SVG/vector icon set, logo system, or illustration library inside the repo\n- Creating simple shapes, diagrams, wireframes, or icons that are better produced directly in SVG, HTML/CSS, or canvas\n- Making a small project-local asset edit when the source file already exists in an editable native format\n- Any task where the user clearly wants deterministic code-native output instead of a generated bitmap\n\n## Decision tree\n\nThink about two separate questions:\n\n1. **Intent:** is this a new image or an edit of an existing image?\n2. **Execution strategy:** is this one asset or many assets/variants?\n\nIntent:\n- If the user wants to modify an existing image while preserving parts of it, treat the request as **edit**.\n- If the user provides images only as references for style, composition, mood, or subject guidance, treat the request as **generate**.\n- If the user provides no images, treat the request as **generate**.\n\nBuilt-in edit semantics:\n- Built-in edit mode is for images already visible in the conversation context, such as attached images or images generated earlier in the thread.\n- If the user wants to edit a local image file with the built-in tool, first load it with built-in `view_image` tool so the image is visible in the conversation context, then proceed with the built-in edit flow.\n- Do not promise arbitrary filesystem-path editing through the built-in tool.\n- If a local file still needs direct file-path control, masks, or other explicit CLI-only parameters, use the explicit CLI fallback only when the user asks for it.\n- For edits, preserve invariants aggressively and save non-destructively by default.\n\nExecution strategy:\n- In the built-in default path, produce many assets or variants by issuing one `image_gen` call per requested asset or variant.\n- In the CLI fallback path, use the CLI `generate-batch` subcommand only when the user explicitly chose CLI mode and needs many prompts/assets.\n- For many distinct assets, do not use `n` as a substitute for separate prompts. `n` is for variants of one prompt; distinct assets need distinct built-in calls or distinct CLI `generate-batch` jobs.\n\nAssume the user wants a new image unless they clearly ask to change an existing one.\n\n## Workflow\n1. Decide the top-level mode: built-in by default, including transparent-output requests; fallback CLI only if explicitly requested or confirmed.\n2. Decide the intent: `generate` or `edit`.\n3. Decide whether the output is preview-only or meant to be consumed by the current project.\n4. Decide the execution strategy: single asset vs repeated built-in calls vs CLI `generate-batch`.\n5. Collect inputs up front: prompt(s), exact text (verbatim), constraints/avoid list, and any input images.\n6. For every input image, label its role explicitly:\n - reference image\n - edit target\n - supporting insert/style/compositing input\n7. If the edit target is only on the local filesystem and you are staying on the built-in path, inspect it with `view_image` first so the image is available in conversation context.\n8. If the user asked for a photo, illustration, sprite, product image, banner, or other explicitly raster-style asset, use `image_gen` rather than substituting SVG/HTML/CSS placeholders. If the request is for an icon, logo, or UI graphic that should match existing repo-native SVG/vector/code assets, prefer editing those directly instead.\n9. Augment the prompt based on specificity:\n - If the user's prompt is already specific and detailed, normalize it into a clear spec without adding creative requirements.\n - If the user's prompt is generic, add tasteful augmentation only when it materially improves output quality.\n10. Use the built-in `image_gen` tool by default.\n11. For transparent-output requests, ask built-in `image_gen` for a transparent background and preserve the generated alpha channel.\n12. Inspect outputs and validate: subject, style, composition, text accuracy, and invariants/avoid items.\n13. Iterate with a single targeted change, then re-check.\n14. For preview-only work, render the image inline; the underlying file may remain at the default `$CODEX_HOME/generated_images/...` path.\n15. For project-bound work, move or copy the selected artifact into the workspace and update any consuming code or references. Never leave a project-referenced asset only at the default `$CODEX_HOME/generated_images/...` path.\n16. For batches or multi-asset requests, persist every requested deliverable final in the workspace unless the user explicitly asked to keep outputs preview-only. Discarded variants do not need to be kept unless requested.\n17. If the user explicitly chooses or confirms the CLI fallback, then use the fallback-only docs for model, quality, size, `input_fidelity`, masks, output format, output paths, and network setup.\n18. Always report the final saved path(s) for any workspace-bound asset(s), plus the final prompt or prompt set and whether the built-in tool or fallback CLI mode was used.\n\n## Transparent image requests\n\nAsk built-in `image_gen` for a genuinely transparent background and preserve its alpha.\n\n## Prompt augmentation\n\nReformat user prompts into a structured, production-oriented spec. Make the user's goal clearer and more actionable, but do not blindly add detail.\n\nTreat this as prompt-shaping guidance, not a closed schema. Use only the lines that help, and add a short extra labeled line when it materially improves clarity.\n\n### Specificity policy\n\nUse the user's prompt specificity to decide how much augmentation is appropriate:\n\n- If the prompt is already specific and detailed, preserve that specificity and only normalize/structure it.\n- If the prompt is generic, you may add tasteful augmentation when it will materially improve the result.\n\nAllowed augmentations:\n- composition or framing hints\n- polish level or intended-use hints\n- practical layout guidance\n- reasonable scene concreteness that supports the stated request\n\nNot allowed augmentations:\n- extra characters or objects that are not implied by the request\n- brand names, slogans, palettes, or narrative beats that are not implied\n- arbitrary side-specific placement unless the surrounding layout supports it\n\n## Use-case taxonomy (exact slugs)\n\nClassify each request into one of these buckets and keep the slug consistent across prompts and references.\n\nGenerate:\n- photorealistic-natural — candid/editorial lifestyle scenes with real texture and natural lighting.\n- product-mockup — product/packaging shots, catalog imagery, merch concepts.\n- ui-mockup — app/web interface mockups and wireframes; specify the desired fidelity.\n- infographic-diagram — diagrams/infographics with structured layout and text.\n- scientific-educational — classroom explainers, scientific diagrams, and learning visuals with required labels and accuracy constraints.\n- ads-marketing — campaign concepts and ad creatives with audience, brand position, scene, and exact tagline/copy.\n- productivity-visual — slide, chart, workflow, and data-heavy business visuals.\n- logo-brand — logo/mark exploration, vector-friendly.\n- illustration-story — comics, children’s book art, narrative scenes.\n- stylized-concept — style-driven concept art, 3D/stylized renders.\n- historical-scene — period-accurate/world-knowledge scenes.\n\nEdit:\n- text-localization — translate/replace in-image text, preserve layout.\n- identity-preserve — try-on, person-in-scene; lock face/body/pose.\n- precise-object-edit — remove/replace a specific element (including interior swaps).\n- lighting-weather — time-of-day/season/atmosphere changes only.\n- background-extraction — transparent background / clean cutout. Ask built-in `image_gen` for actual transparency.\n- style-transfer — apply reference style while changing subject/scene.\n- compositing — multi-image insert/merge with matched lighting/perspective.\n- sketch-to-render — drawing/line art to photoreal render.\n\n## Shared prompt schema\n\nUse the following labeled spec as shared prompt scaffolding for both top-level modes:\n\n```text\nUse case: <taxonomy slug>\nAsset type: <where the asset will be used>\nPrimary request: <user's main prompt>\nInput images: <Image 1: role; Image 2: role> (optional)\nScene/backdrop: <environment>\nSubject: <main subject>\nStyle/medium: <photo/illustration/3D/etc>\nComposition/framing: <wide/close/top-down; placement>\nLighting/mood: <lighting + mood>\nColor palette: <palette notes>\nMaterials/textures: <surface details>\nText (verbatim): \"<exact text>\"\nConstraints: <must keep/must avoid>\nAvoid: <negative constraints>\n```\n\nNotes:\n- `Asset type` and `Input images` are prompt scaffolding, not dedicated CLI flags.\n- `Scene/backdrop` refers to the visual setting. It is not the same as the fallback CLI `background` parameter, which controls output transparency behavior.\n- Fallback-only execution notes such as `Quality:`, `Input fidelity:`, masks, output format, and output paths belong in the CLI path only. Do not treat them as built-in `image_gen` tool arguments.\n\nAugmentation rules:\n- Keep it short.\n- Add only the details needed to improve the prompt materially.\n- For edits, explicitly list invariants (`change only X; keep Y unchanged`).\n- If any critical detail is missing and blocks success, ask a question; otherwise proceed.\n\n## Examples\n\n### Generation example (hero image)\n```text\nUse case: product-mockup\nAsset type: landing page hero\nPrimary request: a minimal hero image of a ceramic coffee mug\nStyle/medium: clean product photography\nComposition/framing: wide composition with usable negative space for page copy if needed\nLighting/mood: soft studio lighting\nConstraints: no logos, no text, no watermark\n```\n\n### Edit example (invariants)\n```text\nUse case: precise-object-edit\nAsset type: product photo background replacement\nPrimary request: replace only the background with a warm sunset gradient\nConstraints: change only the background; keep the product and its edges unchanged; no text; no watermark\n```\n\n## Prompting best practices\n- Structure prompt as scene/backdrop -> subject -> details -> constraints.\n- Include intended use (ad, UI mock, infographic) to set the mode and polish level.\n- Use camera/composition language for photorealism.\n- Only use SVG/vector stand-ins when the user explicitly asked for vector output or a non-image placeholder.\n- Quote exact text and specify typography + placement.\n- For tricky words, spell them letter-by-letter and require verbatim rendering.\n- For multi-image inputs, reference images by index and describe how they should be used.\n- For edits, repeat invariants every iteration to reduce drift.\n- Iterate with single-change follow-ups.\n- If the prompt is generic, add only the extra detail that will materially help.\n- If the prompt is already detailed, normalize it instead of expanding it.\n- For CLI fallback only, see `references/cli.md` and `references/image-api.md` for model, `quality`, `input_fidelity`, masks, output format, and output-path guidance.\n- For transparent images, ask built-in `image_gen` for actual transparency and preserve its alpha.\n\nMore principles shared by both modes: `references/prompting.md`.\nCopy/paste specs shared by both modes: `references/sample-prompts.md`.\n\n## Guidance by asset type\nAsset-type templates (website assets, game assets, wireframes, logo) are consolidated in `references/sample-prompts.md`.\n\n## gpt-image-2 guidance for CLI fallback\n\nThe fallback CLI defaults to `gpt-image-2`.\n\n- Use `gpt-image-2` for new CLI/API workflows unless the user confirms a different model.\n- CLI `gpt-image-2` does not support `background=transparent`; ask before using `gpt-image-1.5` unless the user explicitly requested that model.\n- `gpt-image-2` always uses high fidelity for image inputs; do not set `input_fidelity` with this model.\n- `gpt-image-2` supports `quality` values `low`, `medium`, `high`, and `auto`.\n- Use `quality low` for fast drafts, thumbnails, and quick iterations. Use `medium`, `high`, or `auto` for final assets, dense text, diagrams, identity-sensitive edits, or high-resolution outputs.\n- Square images are typically fastest to generate. Use `1024x1024` for fast square drafts.\n- If the user asks for 4K-style output, use `3840x2160` for landscape or `2160x3840` for portrait.\n- `gpt-image-2` size may be `auto` or `WIDTHxHEIGHT` if all constraints hold: max edge `<= 3840px`, both edges multiples of `16px`, long-to-short ratio `<= 3:1`, total pixels between `655,360` and `8,294,400`.\n\nPopular `gpt-image-2` sizes:\n- `1024x1024` square\n- `1536x1024` landscape\n- `1024x1536` portrait\n- `2048x2048` 2K square\n- `2048x1152` 2K landscape\n- `3840x2160` 4K landscape\n- `2160x3840` 4K portrait\n- `auto`\n\n## Fallback CLI mode only\n\n### Temp and output conventions\nThese conventions apply only to the CLI fallback. They do not describe built-in `image_gen` output behavior.\n- Use `tmp/imagegen/` for intermediate files (for example JSONL batches); delete them when done.\n- Write final artifacts under `output/imagegen/`.\n- Use `--out` or `--out-dir` to control output paths; keep filenames stable and descriptive.\n\n### Dependencies\nPrefer `uv` for dependency management in this repo.\n\nRequired Python package:\n```bash\nuv pip install openai\n```\n\nOptional for image inspection and downscaling:\n```bash\nuv pip install pillow\n```\n\nPortability note:\n- If you are using the installed skill outside this repo, install dependencies into that environment with its package manager.\n- In uv-managed environments, `uv pip install ...` remains the preferred path.\n\n### Environment\n- `OPENAI_API_KEY` must be set for live API calls.\n- Do not ask the user for `OPENAI_API_KEY` when using the built-in `image_gen` tool.\n- Never ask the user to paste the full key in chat. Ask them to set it locally and confirm when ready.\n\nIf the key is missing, give the user these steps:\n1. Create an API key in the OpenAI platform UI: https://platform.openai.com/api-keys\n2. Set `OPENAI_API_KEY` as an environment variable in their system.\n3. Offer to guide them through setting the environment variable for their OS/shell if needed.\n\nIf installation is not possible in this environment, tell the user which dependency is missing and how to install it into their active environment.\n\n### Script-mode notes\n- CLI commands + examples: `references/cli.md`\n- API parameter quick reference: `references/image-api.md`\n- Network approvals / sandbox settings for CLI mode: `references/codex-network.md`\n\n## Reference map\n- `references/prompting.md`: shared prompting principles for both modes.\n- `references/sample-prompts.md`: shared copy/paste prompt recipes for both modes.\n- `references/cli.md`: fallback-only CLI usage via `scripts/image_gen.py`.\n- `references/image-api.md`: fallback-only API/CLI parameter reference.\n- `references/codex-network.md`: fallback-only network/sandbox troubleshooting for CLI mode.\n- `scripts/image_gen.py`: fallback-only CLI implementation. Use only when the user explicitly chooses or confirms CLI mode.\n"}}exectext(ALL_TOOLS.filter(x=>/openai|browser|artifact|publish/i.test(x.name)));
input
text(ALL_TOOLS.filter(x=>/openai|browser|artifact|publish/i.test(x.name)));
result, 4 lines
Script completed Wall time 0.0 seconds Output: []
web__runweb__run
input
const r=await tools.web__run({search_query:[{q:"Codex agent builds tests reviews software",domains:["developers.openai.com","platform.openai.com","learn.chatgpt.com"]}],response_length:"short"});text(r);result, 184 lines
Script completed Wall time 3.5 seconds Output: Best practices | ChatGPT Learn (https://learn.chatgpt.com/docs/learn/best-practices) citeturn0search0 [wordlim: 200] Crawled: today; * Build, test, and lint commands ... It’s a great starting point, but you should edit the result to match how your team actually builds, tests, reviews, and ships code. ... Approval mode determines when Codex asks for your permission to run a command and sandbox mode determines if Codex can read or write in the directory and what files the agent can access. ... You can enable automatic reviews or have Codex reactively review when you @Codex. If you’re new to coding agents, start with the default permissions. Keep approval and sandboxing tight by default, then loosen permissions only for trusted repos or specific workflows once the need is clear. Note that the CLI, IDE extension, and ChatGPT desktop app all share the same configuration layers. Learn more on the sample configuration page. Configure Codex for your real environment early. Many quality issues are really setup issues, like the wrong working directory, missing write access, wrong model defaults, or missing tools and connectors. ## Improve reliability with testing and review Don’t stop at asking Codex to make a change. Ask it to create tests when needed, run the relevant checks, confirm the result, and review the work before you accept it. Codex can do this loop for you, but only if it knows what “good” looks like. That guidance can come from either the prompt or `AGENTS.md`. That can include: * Writing or updating tests for the change * Running the right test suites * Checking lint, formatting, or type checks * Confirming the final behavior matches the request * Reviewing the diff for bugs, regressions, or risky patterns Toggle the diff panel in the ChatGPT desktop app to directly review changes locally. Click on a specific row to provide feedback that gets fed as context to the next Codex turn. A useful option here is the slash command `/review`, which gives you a few ways to review code: * Review against a base branch for PR-style review * Review uncommitted changes * Review a commit * Use custom review instructions If you and your team have a `code_review.md` file and reference it from `AGENTS.md`, Codex can follow that guidance during review as well. This is a strong pattern for teams that want review behavior to stay consistent across repositories and contributors. Codex shouldn’t just generate code. With the right instructions, it can also help test it, check it, and review it. If you use GitHub Cloud, you can set up Codex to run code reviews for your PRs. -------------------------------------------------------------------------------- Maintaining software with the GPT-6 family | OpenAI Developers (https://learn.chatgpt.com/blog/maintaining-software-with-the-gpt-6-family) citeturn0search1 [wordlim: 200] Published: last month; Crawled: today; I work on OpenAI’s Core Experimentation team, where we build and maintain the software that supports experiments across the company, from front-end updates to improvements in model-serving efficiency. ... In this post, I’ll share how we use these models to find cleanup opportunities, gather evidence, and prepare changes for review. ... With GPT-5.6 Sol in Codex, I could specify an area to investigate and clear validation criteria, then let the model write static analysis and remove dead code. ... Code reviews were fanned out to various teams with context on each affected project. ... For example, one service’s container image shrank from 3.31GB to 1.72GB, a 48% reduction, and another set of cleanups removed about 400 visual snapshot variants from our test suite. Oct 14, 2026 Codex # Maintaining software with the GPT-6 family How OpenAI’s Core Experimentation team uses Codex to remove unused code, clean up stale experiments, and improve CI. Author: Rodrigo Roim Image: Maintaining software with the GPT-6 family Every growing codebase needs maintenance: unused code to remove, completed experiments to clean up, and tests to fix. At OpenAI, coding agents have helped us build faster, which has also increased our maintenance needs. I work on OpenAI’s Core Experimentation team, where we build and maintain the software that supports experiments across the company, from front-end updates to improvements in model-serving efficiency. The release of GPT-6 Astra helped us build and maintain software faster, and GPT-6.1 Sol delivered similar results at a lower cost. In this post, I’ll share how we use these models to find cleanup opportunities, gather evidence, and prepare changes for review. ## Removing unused code Before agentic coding, many software engineering teams relied on complex static analysis systems integrated with runtime logging to detect and remove unused code. -------------------------------------------------------------------------------- Build iterative repair loops with Codex (https://developers.openai.com/cookbook/examples/codex/build_iterative_repair_loops_with_codex?trk=article-ssr-frontend-pulse_little-text-block) citeturn0search2 [wordlim: 200] Published: 6 months ago; Crawled: today; We’ll build this agent loop with Codex. Codex reviews the current state, applies focused changes, runs validation, and repeats when the feedback shows remaining issues. ... Codex is no longer working only from the original review; it also sees what happened during validation. May 11, 2026 # Build iterative repair loops with Codex Shreekant Agrawal (OpenAI) View on GitHub Download raw This cookbook is about closed-loop agent workflows: agents that produce an output, validate it, and use the feedback to improve the next pass. We’ll explore a documentation reliability workflow that detects, repairs, and validates stale or broken API and SDK examples. The worked example uses intentionally stale notebooks adapted from this Cookbook repository. We’ll build this agent loop with Codex. Codex reviews the current state, applies focused changes, runs validation, and repeats when the feedback shows remaining issues. The notebook task is only the example. The pattern applies wherever agent output can be measured with trustworthy feedback. The workflow has three phases: * Review: inspect the current artifact and return structured findings without editing files. * Repair: apply focused edits to a copied artifact using the findings and the latest validation feedback. * Validate: run the relevant checks and report what still needs work. Validation closes the loop. The repaired notebook has to satisfy the checks that matter, and any remaining issues become the next repair input. -------------------------------------------------------------------------------- Best practices | ChatGPT Learn (https://learn.chatgpt.com/guides/best-practices?translationFallback=ja-JP) citeturn0search3 [wordlim: 200] Crawled: today; * Build, test, and lint commands ... It’s a great starting point, but you should edit the result to match how your team actually builds, tests, reviews, and ships code. ... Approval mode determines when Codex asks for your permission to run a command and sandbox mode determines if Codex can read or write in the directory and what files the agent can access. ... You can enable automatic reviews or have Codex reactively review when you @Codex. ## Improve reliability with testing and review Don’t stop at asking Codex to make a change. Ask it to create tests when needed, run the relevant checks, confirm the result, and review the work before you accept it. Codex can do this loop for you, but only if it knows what “good” looks like. That guidance can come from either the prompt or `AGENTS.md`. That can include: * Writing or updating tests for the change * Running the right test suites * Checking lint, formatting, or type checks * Confirming the final behavior matches the request * Reviewing the diff for bugs, regressions, or risky patterns Toggle the diff panel in the ChatGPT desktop app to directly review changes locally. Click on a specific row to provide feedback that gets fed as context to the next Codex turn. A useful option here is the slash command `/review`, which gives you a few ways to review code: * Review against a base branch for PR-style review * Review uncommitted changes * Review a commit * Use custom review instructions -------------------------------------------------------------------------------- Build iterative repair loops with Codex (https://developers.openai.com/cookbook/examples/codex/build_iterative_repair_loops_with_codex) citeturn0search4 [wordlim: 200] Published: 6 months ago; Crawled: today; We’ll build this agent loop with Codex. Codex reviews the current state, applies focused changes, runs validation, and repeats when the feedback shows remaining issues. ... We send the review prompt to `codex exec` with a JSON schema. May 11, 2026 # Build iterative repair loops with Codex Shreekant Agrawal (OpenAI) View on GitHub Download raw This cookbook is about closed-loop agent workflows: agents that produce an output, validate it, and use the feedback to improve the next pass. We’ll explore a documentation reliability workflow that detects, repairs, and validates stale or broken API and SDK examples. The worked example uses intentionally stale notebooks adapted from this Cookbook repository. We’ll build this agent loop with Codex. Codex reviews the current state, applies focused changes, runs validation, and repeats when the feedback shows remaining issues. The notebook task is only the example. The pattern applies wherever agent output can be measured with trustworthy feedback. The workflow has three phases: * Review: inspect the current artifact and return structured findings without editing files. * Repair: apply focused edits to a copied artifact using the findings and the latest validation feedback. -------------------------------------------------------------------------------- Meet dots | ChatGPT Learn (https://learn.chatgpt.com/docs/dots?nav=overview) citeturn0search5 [wordlim: 200] Crawled: yesterday; Codex | When you want to build, debug, test, or review software and work directly with code and developer tools. Your dot | When you want an agent to keep track of an ongoing responsibility, figure out next steps as things change, and follow up between conversations. ## How dots compare with ChatGPT Work and Codex Here’s how dots compare with ChatGPT Work and Codex, and how they can work together. Experience | When to use it --- | --- Chat | When you want an answer, an explanation, ideas, or help drafting something in conversation. ChatGPT Work | When you have a task to delegate, such as research, analysis, or creating a file, either now or on a schedule. Codex | When you want to build, debug, test, or review software and work directly with code and developer tools. Your dot | When you want an agent to keep track of an ongoing responsibility, figure out next steps as things change, and follow up between conversations. It can keep you updated in ChatGPT, Slack, or Teams and delegate tasks to ChatGPT Work or Codex. ## Work with your dot Start with work you regularly need to check or update. Describe the responsibility and share the source material it needs, then review its work and explain what needs to change. ### Give it a responsibility For an offsite, share the plan and venue emails, then ask your dot to identify upcoming decisions and deadlines. -------------------------------------------------------------------------------- Codex | OpenAI Developers (https://developers.openai.com/learn/codex) citeturn0search6 [wordlim: 200] Crawled: today; Follow this guide closely to ensure you’re getting th cookbook Introducing the Codex app See the Codex app in action and how it helps you build and ship faster.video Codex in JetBrains IDEs How to use Codex inside JetBrains IDEs like Rider, IntelliJ, PyCharm, and WebStorm.video Modernizing your Codebase with Codex Cookbook to modernize legacy codebases using the OpenAI Codex CLI.cookbook Codex code review Walkthrough of how Codex drives end-to-end pull request reviews with the new onboarding flow.video Build beautiful frontends with OpenAI Codex Learn how OpenAI Codex's multimodal abilities accelerate frontend development.video Context Engineering & Coding Agents with Cursor Session on structuring context for agent workflows inside the Cursor editor.video OpenAI Codex in your code editor Walkthrough of the Codex IDE extension for VS Code, Cursor, and other forks.video Shipping with Codex DevDay talk on building, testing, and delivering products with Codex.video Sora, ImageGen, and Codex: The Next Wave of Creative Production Panel discussion on combining Sora, ImageGen, and Codex for media creation.video Using OpenAI Codex CLI with GPT-5-Codex Overview of running the Codex CLI locally with GPT-5-Codex.video Codex intro Introductory video introducing Codex and its capabilities.video # Codex OpenAI Developers plugin Install and use the OpenAI Developers plugin in Codex, Claude Code, or Cursor.guideCodex for (almost) everything See the latest Codex app updates for working across more of the software development lifecycle.videoCodex Prompting Guide Codex models advance the frontier of intelligence and efficiency and our recommended agentic coding model. Follow this guide closely to ensure you’re getting th cookbookIntroducing the Codex app See the Codex app in action and how it helps you build and ship faster.videoCodex in JetBrains IDEs How to use Codex inside JetBrains IDEs like Rider, IntelliJ, PyCharm, and WebStorm.videoModernizing your Codebase with Codex Cookbook to modernize legacy codebases using the OpenAI Codex CLI.cookbookCodex code review Walkthrough of how Codex drives end-to-end pull request reviews with the new onboarding flow.videoBuild beautiful frontends with OpenAI Codex Learn how OpenAI Codex's multimodal abilities accelerate frontend development.videoContext Engineering & Coding Agents with Cursor Session on structuring context for agent workflows inside the Cursor editor.videoOpenAI Codex in your code editor Walkthrough of the Codex IDE extension for VS Code, Cursor, and other forks.videoShipping with Codex DevDay talk on building, testing, and delivering products with Codex.videoSora, ImageGen, and Codex: The Next Wave of Creative Production Panel discussion on combining Sora, ImageGen, and Codex for media creation.videoUsing OpenAI Codex CLI with GPT-5-Codex Overview of running the Codex CLI locally with GPT-5-Codex.videoCodex intro Introductory video introducing Codex and its capabilities.video-------------------------------------------------------------------------------- Run long horizon tasks with Codex | OpenAI Developers (https://developers.openai.com/blog/run-long-horizon-tasks-with-codex?trk=article-ssr-frontend-pulse_little-text-block) citeturn0search7 [wordlim: 200] Published: 8 months ago; Crawled: today; We recently published an article about the Codex agent loop that has more details. To top this off, we also launched the Codex app that makes that loop usable day-to-day: ... * Git worktrees (isolate runs, keep diffs reviewable, reduce thrash) ... I picked a design tool for this “experiment” because it’s an unforgiving test: UI + data model + editing operations + lots of edge cases. ... Purpose: Freeze the target so the agent doesn’t “build something impressive but wrong.” -------------------------------------------------------------------------------- Automating repetitive work at OpenAI with Codex (https://developers.openai.com/blog/automating-repetitive-work-at-openai-with-codex) citeturn0search8 [wordlim: 200] Published: 2 months ago; Crawled: 4 days ago; I still build software—or, more accurately, Codex helps me build it—but instead of creating a separate automation for every task, I’m building Runme to: ... Google Drive can index that file, which makes previous notebooks easier for an agent to discover when it needs examples, operational context, or the outcome of an earlier run. ... A persistent goal keeps Codex focused on the task, while automatic approval review can review eligible actions without changing the existing permission boundaries. -------------------------------------------------------------------------------- Codex GitHub Action | ChatGPT Learn (https://learn.chatgpt.com/docs/github-action) citeturn0search9 [wordlim: 200] Crawled: today; Use the Codex GitHub Action (`openai/codex-action@v1`) to run Codex in CI/CD jobs, apply patches, or post reviews from a GitHub Actions workflow. ... Replace `.github/codex/prompts/review.md` with your own prompt file or use the `prompt` input for inline text. ... * `model` and `effort`: Pick the Codex agent configuration you want; leave empty for defaults. -------------------------------------------------------------------------------- Codex as a platform: build on the open agent harness | OpenAI Developers (https://developers.openai.com/blog/codex-as-a-platform) citeturn0search10 [wordlim: 200] Published: 3 months ago; Crawled: today; Building on Codex does not require the same integration for every use case. ... The most interesting opportunity is not to reproduce the Codex app with a different logo, but to build software that reflects how a specific person or team already works: ... It tells the agent what the user is looking at, gives it the right tools, and gives the user a place to review what happens next. ... If you want to build with the Codex harness, start with the open-source Codex repository, then choose the integration that fits your product: codex exec for noninteractive jobs, the Codex SDK for programmatic agent workflows, or Codex app-server for applications that need persistent conversations, streamed events, and approval handling. -------------------------------------------------------------------------------- Subagents | ChatGPT Learn (https://learn.chatgpt.com/docs/agent-configuration/subagents?surface=app&trk=article-ssr-frontend-pulse_little-text-block) citeturn0search11 [wordlim: 200] Crawled: today; Subagent activity appears in the ChatGPT desktop app, Codex CLI, and the IDE extension. ... Spawn one subagent for security risks, one for test gaps, and one for maintainability. ... Before applying the file, Codex resolves each setting from an explicit spawn value, then the corresponding `[agents]` default, then the parent’s value. ... Review code like an owner.