Project Repository Setup

Project Repository Setup

Use this skill when creating, reorganizing, or reviewing a software repository's baseline structure and developer workflow, independent of any one programming language or domain.

Repository AGENTS.md instructions take precedence over this skill.

Purpose

Apply a pragmatic default for:

• top-level project layout and repo-root hygiene;
• baseline developer documentation and metadata files;
• pre-commit, formatting, and linting entry points;
• consistent local commands for build, test, and quality checks;
• keeping generated files, caches, and large artifacts out of source layout.

When To Use

Use this skill when:

• the user asks to scaffold or clean up repository structure;
• a project has ad hoc scripts, scattered generated outputs, or unclear entry

points;

• baseline files such as .gitignore, .editorconfig, README, or

CONTRIBUTING need to be added or rationalized;

• pre-commit hooks or standard local validation commands should be introduced;
• CI and local developer workflows should use the same project entry points.

Do not use this skill to override a framework's required layout or an existing repository convention that is already intentional and working.

Working Approach

When this skill applies:

1. Read the repository AGENTS.md first, if present.
2. Inspect existing top-level directories, scripts, and generated outputs.
3. Identify which project files are source, tests, docs, automation, examples,

or generated artifacts.

1. Prefer small, conventional layout improvements before large migrations.
2. Keep local developer commands and CI entry points aligned.
3. State what was reorganized, what was left unchanged, and what follow-up

cleanup remains.

Prefer incremental structure changes unless the user explicitly asks for a broad repository reorganization.

Top-Level Layout

A practical language-agnostic baseline is:

• src/ or a project-specific source directory for primary implementation code;
• tests/ for automated tests;
• docs/ for user or developer documentation beyond the README;
• examples/ for runnable examples or small sample projects;
• scripts/ for maintenance or developer automation;
• benchmarks/ for benchmark drivers or performance experiments when relevant.

Only create directories that have a clear near-term use. Avoid empty layout theater.

Repo Root Hygiene

Keep the repository root focused on discoverability:

• README.md for project purpose and common workflows;
• LICENSE or equivalent licensing file;
• CONTRIBUTING.md when contribution workflow matters;
• .gitignore for generated artifacts, caches, build trees, and local-only files;
• .editorconfig when basic editor consistency is useful;
• AGENTS.md when future Codex work should follow repository-specific

conventions or use domain skills;

• one or a few project-level build/config entry points.

Avoid dumping generated outputs, one-off scratch files, or many unrelated helper scripts into the root.

Agent Instructions

For new scientific-computing repositories, add a concise root AGENTS.md unless the user declines or the repository already has equivalent agent guidance. The file should:

• describe the repository scope and any source, test, docs, examples, and

generated-artifact boundaries;

• point Codex agents at this scientific-computing skill set for domain work,

using the relevant scientific-* skills rather than generic software advice;

• list the repository's normal validation commands once they exist;
• avoid site-specific HPC login, scheduler, filesystem, account, allocation, or

cluster-host facts unless the target repository is explicitly site-local.

Keep AGENTS.md short and stable. Prefer project conventions and validation commands over broad instructions that duplicate every skill.

Pre-Commit and Quality Hooks

• Use pre-commit hooks to run fast, deterministic checks that catch common

formatting and hygiene issues before commit.

• Keep hooks fast enough that developers do not routinely bypass them.
• Prefer repository-local config files and pinned tool versions where practical.
• Avoid hooks that depend on machine-local state or hidden external services.
• Make hook behavior consistent with CI checks so the same issues are caught in

both places.

Treat expensive tests, flaky integration checks, or environment-specific validation as poor pre-commit defaults.

Use developer-tool-installation-policy when deciding whether tools used by hooks, local commands, or CI should be project-pinned or installed as isolated global CLIs.

Local Developer Commands

• Provide one obvious command path for setup, test, format, and lint when the

project supports those actions.

• Prefer checked-in scripts, task runners, or documented commands over ad hoc

shell snippets hidden in docs.

• Keep command names and behavior stable enough for CI and contributors.
• Avoid duplicating near-identical logic across local scripts and CI workflows.

If the repository spans multiple tools or languages, make the top-level command surface clear even if internals remain tool-specific.

Generated Files and Artifacts

• Keep build directories, caches, temporary outputs, and generated artifacts out

of source directories when practical.

• Add ignore rules for files that are reproducible and not meant for review.
• Commit generated files only when there is a clear reason, such as vendored

lockfiles, generated documentation intended for publication, or checked-in golden references.

• Store large data artifacts outside the normal source tree unless they are

truly part of the repository contract.

Repository structure should make it obvious what humans edit and what tools produce.

Branch and Commit Hygiene

• Document repository expectations for single-purpose branches and atomic

commits when contributors or agents regularly prepare pull requests.

• Encourage agents to inspect git status --short before staging and to warn

when unrelated dirty files would pollute a commit or pull request.

• Use git-refactor-hygiene for move/rename work and for branch, staging,

commit, or pull-request scope checks.

Configuration Files

• Place tool configuration where the tool and repository conventions expect it.
• Prefer one canonical config per tool rather than scattered duplicates.
• Keep shared settings in version-controlled files and local overrides in

ignored machine-specific files.

• Document non-obvious config interactions in the README or contribution docs.

Avoid configuration sprawl that makes it unclear which file actually controls a tool.

CI Alignment

• Reuse the same scripts, presets, or task-runner targets locally and in CI when

practical.

• Keep CI jobs pointed at documented project entry points rather than bespoke

one-off command sequences.

• Make pre-commit, lint, test, and build expectations visible from the repo

itself, not only from CI logs.

Local and CI workflows should validate the same contract unless there is an explicit reason to split them.

For GitHub-hosted projects, GitHub Actions is a practical default CI target. Keep workflow files under .github/workflows/, run documented local commands inside jobs, avoid broad permissions, and keep release or deployment automation separate from ordinary validation unless the repository deliberately uses CI/CD.

Use scientific-ci for scientific CI design details such as build matrices, numerical validation, artifacts, GitHub Actions job structure, and portability tradeoffs.

Anti-Patterns

• creating many empty directories with no current purpose;
• mixing source files, generated artifacts, and scratch outputs in one tree;
• maintaining separate undocumented commands for local development and CI;
• adding pre-commit hooks so slow or flaky that they get ignored;
• hardcoding machine-local paths in repository automation;
• moving large parts of the tree purely for aesthetics without improving

discoverability or workflow.

Validation Defaults

• Check that moved or newly referenced scripts and config files exist.
• Run the lightweight local quality or test entry point when practical.
• Verify that ignore rules do not hide source files that should be tracked.
• State explicitly which setup, hook, or validation paths were exercised.
• If a proposed layout change was not applied, note the reason and migration

risk.

Output Expectations

When using this skill, briefly note:

• which top-level layout, metadata, or automation files were added or changed;
• whether pre-commit, lint, format, or test entry points were introduced;
• how generated artifacts and local-only files are handled;
• which local validation commands were run;
• which repository-structure tradeoffs or follow-up cleanup remain.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: gjbex
• Source: gjbex/scientific-computing-skills
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.