# openprompt.co: a daily table that ranks jailbreak prompts beside translation prompts

> openprompt.co regenerates a star-ranked list of the most saved prompts every 24 hours and commits it back as markdown. The pipeline is three dependencies and one script, and the list it produces places a small number of guardrail-evasion prompts in the same table as translators and code reviewers.

**timqian/openprompt.co** — Create. Use. Share. ChatGPT prompts

- Repository: https://github.com/timqian/openprompt.co
- Website: https://openprompt.co
- Stars: 1,216 · Forks: 59
- Language: JavaScript
- License: GPL-3.0
- Published: 2026-09-14 · Updated: 2026-09-14 · Language: en
- Canonical page: https://hysenlabs.com/projects/timqian-openprompt-co

## The repository is a generated table, not an application

The whole system is one script and one data file.

Top-level entries are a GitHub directory, a gitignore, a licence, this README, a JSON file called TopPrompts.json, a script called `index.mjs`, a package manifest and a yarn lockfile. There is no server, no application directory and no test suite.

```json
{
  "name": "openprompt.co",
  "version": "0.0.0",
  "main": "index.mjs",
  "license": "GPL-v3",
  "dependencies": {
    "@supabase/supabase-js": "^2.12.0",
    "dotenv": "^16.0.3",
    "json-to-markdown-table": "^1.0.0"
  }
}
```

Three dependencies tell you the entire pipeline. A Supabase client to read the prompt records, dotenv to pick up credentials, and a library whose only job is turning JSON into a markdown table.

So the README you are reading is output. Once every 24 hours the script queries the database, renders ranked entries with a link, an author page and a star count, and writes the result over the previous README. The commit history of that file is a daily archive of a leaderboard.

That design has one obvious virtue and one obvious cost. The virtue is that the data is reproducible and diffable. The cost is that the repository's only content is a snapshot of somebody else's database, and the table cannot be corrected by editing the file.

## One contributor holds six of the visible twenty entries

The list is presented as a community ranking, so it is worth counting who is actually in it.

One account, shown as `tim`, holds six of the twenty entries on the visible page: the top slot with 808 stars, the translate-to-Chinese entry at 255, refactoring at 80, the six-person advisory board at 73, the code-lengthening prompt at 64, and the profanity persona at 58. A second account, `timqian`, appears on the English translator entry at 135, which is plausibly the same person under a different handle.

Beyond that, one account holds two entries, a Chinese polisher at 67 and a text de-duplication prompt at 49. And one of the attribution links is a bare numeric string rather than a username, which reads like a generated identifier rather than a person.

So of roughly twenty visible entries, one contributor accounts for about a third, including the highest ranked one, and the remainder are single contributions spread across handles that look like platform user identifiers rather than chosen names.

That is not a criticism of the directory, which never claims editorial curation. It is a reason to read the star column as a measure of early accumulation rather than of a considered shortlist, and to treat the table as a popularity record.

## Three of the twenty are guardrail-evasion or evasion-of-detection prompts

This is the part a reader needs to know before pointing an agent at the list.

Three entries in the visible top twenty are not ordinary task prompts.

The entry at number eighteen is a jailbreak in the classic style. It opens by telling the model to ignore its previous instructions, asserts a persona that is supposed to have no content policy, and instructs the model to answer every message twice, once normally and once in the uncensored mode, with tags in front of each. It closes with a threat that failure to comply will result in the account being disabled.

The entry at number fourteen is a persona prompt that instructs the model to always reply with profanity, to keep doing so when criticised or subjected to moral judgement, and to conceal that it is an assistant at all. When it cannot answer a question it is told to reply with profanity anyway.

The entry at number seventeen describes itself as a de-duplication tool: its stated job is to reorganise text by changing wording and adjusting word order so that the rate of repetition in the output is as low as possible, without changing the meaning. That is a description of evading similarity detection, which is what plagiarism checkers look for.

All three sit in the same numbered, star-ranked table as the translators and the code reviewer. The directory aggregates what people saved rather than approving it, and none of these three is flagged, dated as unsafe, or separated from the rest.

## Four of the twenty are translation and language polishing prompts

The crowded category is language work, and the crowd is not accidental.

Rank two translates any language into Chinese and, when the input is already Chinese, polishes it. Rank four is an English translator, spelling corrector and improver that asks for literary register. Rank eleven polishes Chinese sentences into formal written language and fixes punctuation. Rank thirteen translates into English and flags awkward constructions in brackets.

Four of twenty for one task family, and the descriptions are written in Chinese. The site is openprompt.co with a `.co` domain, its links are language-neutral slugs, and the visible leaderboard is dominated by Chinese-language prompts with Chinese titles.

The two code entries follow the same pattern. Rank seven is a request to review code, rank eight to refactor for efficiency, and rank twelve asks for code length to be multiplied five times without changing behaviour, with tidier variable names and added comments. Three of twenty for code, two of them by the same contributor who holds the top slot.

The practical reading is that this is a Chinese-language prompt directory whose top of table is dominated by translation, code and personality prompts written for one audience, with the highest star counts accumulated during the first months of ChatGPT.

## Version 0.0.0, a licence string that is not SPDX, and no releases

The packaging metadata has three small problems worth naming.

```json
  "version": "0.0.0",
  "repository": "git@github.com:timqian/openprompt.co.git",
  "license": "GPL-v3",
```

The version has never been set. For a package that is not published and not installed, that is defensible, but it also means there is no way to refer to a state of the code by version.

The licence string is `GPL-v3`. The SPDX identifier for the GNU General Public License version 3 is `GPL-3.0`, so this string will not match anything that validates licences, including automated tooling. The project's own licence field records GPL-3.0, and a LICENSE file is present, so the intended terms are unambiguous; only the manifest string is wrong.

The repository field uses the SSH form, `git@github.com:...`, while every link in the README uses https. That is a stylistic inconsistency rather than a fault, and it does mean a clone command copied out of the manifest needs keys.

And there are no GitHub releases at all. For a project whose entire value is a time series, the absence of tags is understandable, since a release would duplicate what the daily commit already records. The last push to the repository is dated 2026-09-24, which is the only freshness signal the page offers.

## The ranking is decided by stars accumulated in 2023

The list is described as the most starred prompts on the site, and the star column runs from 808 at the top to 49 at the bottom of the visible page.

The distribution is steep at the top and crowded at the bottom. Six entries have between 808 and 255 stars. Then there is nothing between 255 and 142. Below that it flattens into a cluster: 135, 120, 99, 82, 80, 75, 73, 67, 64, 59, 58, 56, 49 and 49.

So the top of the table is decided and the tail is a near-tie between entries that differ by a handful of saves. Any reordering in the tail is noise.

More importantly, the metric has an age. The entries describe a product landscape from early ChatGPT: prompts written to simulate a GPT-4 persona, to act as a translator with a specified register, to generate Midjourney prompt sets in the format that tool used at the time. One entry cites a source on Twitter from 2023, and one describes itself as for research only with a request not to use it illegally, which is the disclaimer convention of that moment.

That does not make them useless. It makes the ranking a record of what was popular when the directory started, not a current recommendation, and a reader should treat it accordingly.

## TopPrompts.json is the part to consume, not the README

The page says twice that the same data is available in JSON form, at TopPrompts.json in the repository root, and it invites issues through the GitHub issue tracker for suggestions about the site.

That JSON file is the useful artefact. It is the machine-readable form of the table, it lives in version control, and it changes once a day. Anything that wants the ranking rather than the prose should read that file, because parsing a markdown table is strictly worse work than parsing the object it was rendered from.

The site itself is the interactive half, and the README's role is to advertise it: every entry links out to a page on openprompt.co, and every author links to a profile page there, which is where the prompts actually live.

So there are three surfaces with different purposes. The site is where prompts are submitted and saved. The repository holds the daily ranked snapshot and its history. And the JSON is the interface for anyone building on top of the directory.

One thing the repository does not do is explain how entries get into the ranking, whether stars can be gamed, whether prompts are reviewed before appearing, or whether the three evasion entries were ever flagged. The page is a leaderboard, and a leaderboard answers the question it was built to answer and nothing beyond it.

## Conclusion

openprompt.co is worth using as a historical artefact rather than a reference: it is a versioned, dated snapshot of what people saved in 2023, which is useful precisely because the ranking has not moved. Two cautions if you point an agent at it. The visible list includes prompts whose stated purpose is to defeat model safety policies and one whose purpose is to reduce a text's similarity to its source, and those are presented in the same ranked table as everything else, so treat the directory as unvetted content rather than as a curated collection. And the README is generated output, so editing it by hand accomplishes nothing. If you want the data, take TopPrompts.json; if you want the history, read the commits.

## FAQ

### What is openprompt.co and how often does it update?

It is a star-ranked directory of ChatGPT prompts. The repository README is regenerated every 24 hours from a Supabase database by a single script and committed back, with the same data also published as TopPrompts.json.

### What dependencies does the openprompt.co script use?

Three: the Supabase JavaScript client to read the records, dotenv for configuration, and a JSON-to-markdown-table library that renders the ranked table. The manifest declares the main entry point as `index.mjs`.

### Does openprompt.co review the prompts it lists?

The page describes a star ranking of saved prompts and invites suggestions through GitHub issues. It says nothing about review, moderation or flagging, and three of the twenty visible entries are guardrail-evasion or similarity-evasion prompts presented in the same table.

### Who writes the most prompts in the openprompt.co list?

One account holds six of the twenty visible entries, including the top-ranked one with 808 stars, plus a translate, a refactor, an advisory-board, a code-lengthening and a profanity-persona prompt. Another handle appears on the English translator entry.

### Is openprompt.co still released or versioned?

No releases and no versions. The manifest version is 0.0.0, there are no GitHub releases, and the last push to the repository is dated 2026-09-24. The daily commits are the only version history.

## Sources

- [Issues](https://github.com/timqian/openprompt.co/issues)
- [License: GPL-3.0](https://github.com/timqian/openprompt.co/blob/main/LICENSE)
- [Project website](https://openprompt.co)
- [README](https://github.com/timqian/openprompt.co/blob/main/README.md)
- [timqian/openprompt.co on GitHub](https://github.com/timqian/openprompt.co)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/timqian-openprompt-co