# AI Brief 001 | AI coding assistants in an existing project

Public practice kit v1.0 · 2026-10-30

Practice task scoping, AI collaboration and change acceptance. This is not a comparative product evaluation and contains no scores or prefilled success records. `starter/brief_tasks.py` intentionally retains a timezone/date bug. Status filtering is not implemented yet.

## Start

1. Keep an unmodified original. Copy `starter` to `work` and edit only that copy, or keep a recoverable original with your own version-control tool.
2. Use locally installed Python 3.10 or newer. The command may be `python` on Windows or `python3` on other systems; adapt it to your environment. No third-party dependencies are needed.
3. Read `TASK.md` and `ACCEPTANCE.md`. Fix date filtering first; add status filtering only after accepting the fix.
4. Give the relevant files and a template from `PROMPTS.md` to your chosen assistant. The template requests analysis and a confirmed plan before editing.
5. Run the script in the practice folder and compare the output with the criteria. It reads the sibling `tasks.json` and prints JSON. It does not connect to a network or write files.

Example, after entering the `work` folder:

```console
python brief_tasks.py --on 2026-10-31
```

**Known starter behavior:** this command returns b03 and b04; the intended result is b01 and b03. This describes the teaching code, not a completed run or fix by any AI tool.

## Files

- `TASK.md`: scope, input and output contract, in both languages.
- `ACCEPTANCE.md`: expected results, commands and review checklist, in both languages.
- `PROMPTS.md`: Chinese and English starting prompts.
- `DECISION.md`: workflow choice and reflection sheet, in both languages.
- `runs.csv`: headers only; fill in your actual observations.
- `starter/`: source code and five fixed example records.

## Recording conventions

Record environment setup separately from execution time. `human_interventions` counts interventions to correct interpretation, errors or scope; simply observing does not count. `criteria_passed` and `criteria_total` refer to ACCEPTANCE.md. Put unperformed checks under `not_checked`, never count them as passes. Save actual command output to your own record location and enter it under `evidence_path`.

Start comparisons from the same original copy each time. Record model, tool, permissions, context and date. One small exercise cannot establish an overall product ranking. For a real project, check whether you are permitted to share its code and data with your selected service.

## Permission to use

You may use and modify this kit's original code and text for free personal or internal team practice. When redistributing it, credit AI Brief and retain the explanation that it is teaching material, not a product evaluation. Third-party product documentation is governed by its own terms and is not included in this kit.

Web issue: https://ai-brief-5c6a15.pages.dev/briefs/ai-coding-existing-project.html