Using Cursor and Aider for Bioinformatics

Cursor is a graphical code editor with AI features; Aider is a command-line coding assistant that works with a Git repository. This hands-on lab is for beginners who know how to use a terminal and Git. By the end, you will compare the same one-sentence edit in two disposable copies, inspect the exact diff, and reject anything outside the request. No biological data or analysis is involved, and neither assistant validates scientific reasoning.

Keep the assistant inside a reviewable project

Use only this disposable, documentation-only project, which contains invented arithmetic and no biological data. Download the offline Git diff validator and save it as check_diff.py in the project folder before committing the baseline. Then create the known README and commit both files:

git --version
mkdir cursor-aider-practice
cd cursor-aider-practice
cat > README.md <<'EOF'
# Sample-count helper

This documentation-only project describes a tiny arithmetic example.

Input: 12 reads in sample A and 8 reads in sample B.
Expected total: 20 reads.
EOF
git init
git add README.md check_diff.py
git commit -m "Add project baseline"
git status --short

Run this in a new, writable directory. Git must be configured to make a local commit; if it is not, use Git's local repository identity instructions and a placeholder identity for this disposable fixture rather than changing global settings. The final status should be empty. Stop if it shows unexpected files or changes.

Create two independent copies from that committed baseline so the comparison starts from the same text. From inside cursor-aider-practice, run:

git clone . ../cursor-practice
git clone . ../aider-practice
git -C ../cursor-practice status --short
git -C ../aider-practice status --short

Both status commands should print nothing. Do not put research files, credentials, private outputs, or unrelated project files in either copy. Before each live trial, inspect its status again and stop if it is not clean.

Install Cursor

Download the desktop app from the official Cursor download page. Cursor's current quickstart lists macOS 12 or later, Windows 10 or later, and Linux installation options. Installer formats and supported versions can change, so follow the current official quickstart. Install and sign in only if you choose to try the live comparison and the app requires it; no account is needed for the offline fixture.

1. Open Cursor and choose File → Open Folder, then open the cursor-practice copy.
2. Open Agent and use the exact shared task below. Review the proposed diff before accepting changes into the working tree.
3. Run the offline validator yourself; do not treat the editor's review interface as proof that the change is correct.

Cursor's quickstart describes opening Agent with Cmd+I. Key bindings and labels can change, so consult the quickstart if the shortcut differs. Cursor's data-use documentation describes handling based on privacy settings; it also states that requests go through Cursor's backend even when you use your own API key. Review current settings and terms before sharing any project context.

Install Aider in an isolated environment

For the optional live comparison, Aider's current installation guide recommends its aider-install bootstrap with an existing Python 3.8 to 3.13. It installs Aider in a separate environment and may install a separate Python 3.12. Follow the current official installation guide; do not install it into a carefully managed research environment. The offline fixture does not require Aider.

python -m pip install aider-install
aider-install
aider --version

Optional isolated alternatives include pipx and uv, as described in the current installation guide. Choose one documented method only. Version support and commands can change.

Provider, cost, and data flow are your choice

You may skip provider setup and the model call. If you choose to use Aider, select and configure a provider yourself using its current API-key documentation. Hosted requests may incur charges. Confirm the selected model, current price, provider data terms, and Aider's optional analytics settings before a request. Aider sends files added to its chat and may include related repository context; add only README.md for this fixture. Never put a real key in source code, a prompt, command history, or a shared notebook, and never log or print secrets. An environment variable or ignored local file can still be exposed by the machine or shell.

Make the same bounded edit in both copies

For both tools, use this exact prompt, without adding biological context:

Prompt: Edit only README.md. Immediately after the Input line, add exactly this sentence: "The two sample counts sum to 20 reads." Do not alter any existing text, the values 12, 8, or 20, or any other file; do not add dependencies, inspect or expose secrets, or access data outside this folder.

In Cursor, open Agent in cursor-practice and submit that prompt. In Aider, start from the separate aider-practice copy, add only README.md, and disable automatic commits so the edit stays inspectable as a working-tree change:

cd ../aider-practice
aider --no-auto-commits README.md

Aider's documentation says normal Git integration commits edits automatically; --no-auto-commits disables that behavior for this comparison. If your installed version does not recognize the option, stop and check its current help instead of proceeding with a different workflow.

The only acceptable change is this insertion between the existing Input and Expected total lines:

 Input: 12 reads in sample A and 8 reads in sample B.
+The two sample counts sum to 20 reads.
 Expected total: 20 reads.

python3 check_diff.py

Run project checks and interpret failures

Run tests or validation commands already documented by the project. Do not assume a generic command exists or install new packages just to make an assistant's output appear validated. For scientific code, verify input files, parameters, units, sample identifiers, and outputs against the analysis plan and appropriate controls.

If a command fails, identify whether the cause is a missing dependency, an incorrect path, a code error, or an environment mismatch before asking for a fix. If it passes, record exactly which command ran and on what test data. A passing software test does not validate biological interpretation or establish reproducibility on another platform.

R-first exercise: inspect the synthetic count table

Use the same invented three-gene by two-sample TSV, with a genuine base-R implementation. Download the synthetic TSV, the R checker, and its focused tests into one folder. Run:

Rscript inspect_counts.R counts.tsv
Rscript test_inspect_counts.R inspect_counts.R counts.tsv

Predict the orientation and totals first. The header identifies genes in rows and samples in columns. The output should report sample_A: 16, sample_B: 12, and an arithmetic total of 28. The checker requires a rectangular table, unique nonblank identifiers, and finite nonnegative integer count syntax. Its tests reject negative, fractional, non-finite, ragged, and duplicate-identifier cases. Those checks do not establish provenance, raw-count status, sample independence, or biological validity.

Deliberately faulty proposal: “Each row is a sample, so the sample totals are 20, 3, and 5.” Reject it. Those are row sums for genes because the table is genes by samples, not per-sample totals. Aider installation and the Git diff validator above remain Python-only native tools; no R API or R wrapper is claimed.

Optional Python counterpart for the same count-table check

This exercise uses only an invented three-gene by two-sample table. Download the synthetic TSV and standard-library checker into one folder, then run:

python3 inspect_counts.py counts.tsv

Before running it, predict the orientation and sample totals. The first column is gene identifiers, rows are genes and columns are samples. The checker requires a rectangular table with unique identifiers and finite, nonnegative integers, then sums each sample column. The expected column sums are sample_A: 16 and sample_B: 12; the arithmetic total is 28.

Deliberately faulty proposal: “Each row is a sample, so the sample totals are 20, 3, and 5.” Reject it. Those are row sums for genes because the table is genes by samples, not per-sample totals. The correct response is to report the column sums above and explain the orientation.

Answer rubric: identify genes as rows and samples as columns; report 16 and 12 by summing columns; reject negative values, decimals, NaN, and infinity as not nonnegative integer counts; and state that shape, integer, and sum checks do not prove that values are truly raw counts, have correct provenance, preserve sample identity, represent independent samples, or support any biological conclusion. The example is synthetic arithmetic only. Do not infer biology from it.

Python tool boundary: the optional Python checker uses the Python standard library. Cursor, Aider, hosted providers, and model responses were not installed or run; no model request or paid call was made. Hosted model use is optional and may incur charges.

Choosing between the two tools

Use Cursor when an interactive editor and visual diff review fit your work. Use Aider when a terminal workflow in a Git repository fits better. They can complement each other, but using both does not provide independent scientific validation: both may produce plausible, incorrect changes. Keep analysis design and biological interpretation with the researcher.