| name | autoacad-prepare |
| description | Use when starting a paper project and needing to inventory local code, docs, datasets, constraints, and source materials before literature review or experiment planning. |
AutoAcad Prepare
Overview
Use this stage to establish the paper project's ground truth before any literature or experiment work.
Workflow
- Read the project docs first: abstract draft, experiment design notes, whole-paper concept notes, and any constraints file.
- Inventory code, data, examples, and docs that the paper must describe faithfully.
- Detect hardware constraints early. Record whether the project is CPU-only, CUDA, or Apple MPS.
- Identify missing prerequisites: absent docs, missing dataset paths, missing template, or missing baseline implementations.
- Initialize the canonical project tree with
../scripts/init_paper_project.py if the project is not structured yet.
- Write the first stage plan and the initial
PROGRESS.md entry before moving on.
- For reference codebase selection (if applicable):
- Review searching results and repositories for relevance to innovative ideas.
- Selection criteria:
- Repositories with more stars are more recommended.
- Repositories created more recently are more recommended; too old repositories are not recommended.
- More detailed
README.md file means more readable codebase and more reproducible, so more recommended.
- More clear code structure, code comments, and inline code explanations mean more readable codebase and more maintainable, so more recommended.
- Prefer repositories with
python language, and running coding in the local machine rather than in docker. For deep learning projects, prefer pytorch framework.
- Choose at least 5 repositories as reference codebases, aiming for accuracy and minimal number.
Required Outputs
- A concise source-of-truth inventory.
- A list of blocked inputs and assumptions.
- A confirmed project tree rooted at
<project>-paper/.
- A first-pass environment note in
PROGRESS.md.
- If applicable, a list of determined reference codebases, paths, and papers.
