Prompt library
These are 100 real task descriptions people gave Jaah, grouped by the kind of work they asked for. Use them to see the range of work a task can take on, and as models when you write your own. Each example shows the description as it was written and what the work took.
The descriptions are lightly edited: typos are fixed, long ones are condensed, and the names of people, organizations, projects, hosts and links are removed. Where a description names the product being built it says AppName, and its website reads website.com. The figures were read on 2026-10-06.
Each group below is a page. How to write a description like these is in Write a good task description; how to start one is in Start a task. What a task, an epic and a persona are is in How tasks work, How epics and features work and How personas and playbooks work.
What each example shows
| Figure | Meaning |
|---|---|
| Size | Sketch (a mockup to look at), Task (one agent delivers it) or Epic (analysed, designed and split into subtasks first). |
| Tags | The kind of work, from the list below. |
| Time | From start to finish, including time waiting for a person to answer; then the time agents were actually working. An epic adds up subtasks that ran in parallel, so its agent time can exceed the time from start to finish. |
| Questions asked | Clarifying questions the agents asked a person. |
| Tokens | Model tokens in and out, across the task and all its subtasks. |
| Cost | Approximate model cost at list price, in US dollars. |
| Change | Lines added and deleted across the merged pull requests. |
| Pull requests | How many pull requests merged. |
| Attachments | Files (screenshots, mockups, documents) that came with the description, when there were any. |
Tags
| Tag | Meaning |
|---|---|
| feature | Builds something new. |
| fix | Repairs broken behaviour. |
| investigation | Finds out why something happens, often with a fix. |
| research | Delivers a report or advice, not code. |
| refactor | Restructures without changing behaviour. |
| content | Writes or edits copy, scripts or media. |
| mockup | A visual mockup, no code. |
| frontend | User interface. |
| backend | Server logic and APIs. |
| mobile | Phone-sized screens. |
| data | Calculations, spreadsheets and databases. |
| infra | Cloud, servers and deployment. |
| ci | Build pipelines, tests and branches. |
| security | Access control, compliance and hardening. |
| integration | Connects to an outside service. |
| docs | Documentation and explanations. |
| website | A public website. |
| video | Video scripts and recordings. |
| attachments | The description came with screenshots, mockups or files. |
| robust | Small but powerful: a description of at most 400 characters that produced over 1,000 changed lines or over 100 M tokens of work. |
Groups
- Build features — New capabilities taken from a paragraph of intent to a merged change. (19 examples)
- UI and UX — Layout, polish and behaviour, usually explained with a screenshot. (13 examples)
- Fix and investigate — From a symptom to a root cause and a fix. (11 examples)
- Data, calculations and reports — Formulas, spreadsheet exports and database work. (13 examples)
- Infrastructure, CI and security — Pipelines, cloud deployments and access control. (10 examples)
- Refactoring and maintenance — Keeping a codebase lean and its history clean. (4 examples)
- Research and advice — Answers, analyses and options, delivered as a report rather than code. (9 examples)
- Website content and SEO — Pages, copy and discoverability for a public website. (7 examples)
- Video and visual assets — Tutorial scripts, recorded videos and images. (3 examples)
- Epics — Larger asks that are analysed, designed and split into subtasks before anything is built. (6 examples)
- Sketches — A quick mockup to look at before anything is built. (5 examples)