Use a writing skill.

Copy a prompt

1. Choose a skill

   Search the catalog or pick a category. Each skill page lists its use cases. Technical English is for precise procedures. Clear prose preserves voice. Caveman uses terse fragments.

2. Copy the prompt

   Select Copy prompt above the complete prompt. Paste it into your chat. Add your text or question.

3. Keep what matters

   Review the answer for missing facts, conditions, and uncertainty. Keep all required meaning.

If the browser blocks copying, the page selects the prompt. Copy it with your browser’s Copy command.

Browse the skills

Search the catalog

1. Describe your text in What do you need to shorten? Results update as you type.
2. Or select a suggested task, such as A live outage update. It fills the search field and runs the same search.
3. Select a category to narrow the results. All skills removes the category filter.
4. Review the three best matches. Select Show all matches to see the rest.
5. If no skill matches, keep your words and edit them. Or select Clear search, a suggested task, or Show all 100 skills.

Keyword search needs no download. To add matching by meaning, open Semantic search, read its note, and select Use semantic search. First use downloads about 45 MB: runtime files from jsDelivr and model files from Hugging Face. Your browser processes each query. It sends no query to a model service. If loading fails, keyword search continues.

Search URLs include the query and category. Opening, reloading, or sharing such a URL sends the query to the site host. Without JavaScript, the catalog lists every skill.

For agents

An agent needs web access. It chooses one skill, reads the complete prompt, and applies it to the requested task.

Choose a skill

Start with select.json (JSON). It maps each skill key to a short use case. Replace {id} in instructions_url_template with the chosen key, then fetch that prompt.txt. This needs two fetches.

Read https://sho.rten.it/select.json. Choose the skill that fits my task. Replace {id} in instructions_url_template with its key. Read that complete prompt before you apply it to my text.

Use a known skill

Read /methods/<slug>/prompt.txt for the complete prompt in one fetch. Skip discovery when you know the skill. Caveman’s prompt.txt is an example.

Read the complete prompt at https://sho.rten.it/methods/caveman/prompt.txt. Then apply the Caveman skill to my next request.

Apply the skill

• Use the skill that the user names. Search only when the user names none.
• Read the complete prompt before you apply it. If a fetch fails or returns incomplete text, report that limit. Do not claim to have used the skill.
• Keep the user’s language, format, and required detail. Apply the skill only to the requested writing task.
• If no skill fits, do not force a match. Fetching a prompt does not authorize external actions.

Other formats

• index.json adds names, categories, summaries, and use cases when a selection is unclear.
• catalog.json and llms.txt hold the full catalog as JSON. If one is already in context, choose from its methods.
• Each skill’s SKILL.md adds a descriptive header. Its llms.txt adds the example and sources as JSON.
• Static JSON URLs do not filter results from query parameters. Browser search runs on the visitor’s device, not as an API.
• Public /methods/<slug>/AGENTS.md files are legacy pointers.

Full catalog JSON