precommit-setup

Pre-commit Setup Skill

Configure a three-layer pre-commit quality system that

enforces linting, type checking, and testing before every

commit.

When To Use

• Setting up a new project with code-quality enforcement
• Adding pre-commit hooks to an existing project
• Upgrading from basic linting to a full quality system
• Setting up monorepo or plugin architecture with

per-component quality checks

• Updating pre-commit hook versions

When NOT To Use

• Pre-commit hooks already configured and working optimally
• Project does not use git version control
• Team explicitly avoids pre-commit hooks for workflow reasons

Philosophy: Three-Layer Defense

The system is organised in three layers, each with a

different cost / coverage tradeoff:

• Layer 1: Standard hooks: fast global checks

(50-200ms total). Lints and type-checks every staged file.

• Layer 2: Component-specific checks: per-component

lint, typecheck, and test (10-30s total). Only the

components touched by the staged files are run.

• Layer 3: Validation hooks: project-specific

structure and pattern checks (varies). Catches violations

that generic linters miss.

This layering keeps the fast feedback loop fast while still

catching the slow / project-specific bugs before they land.

Module Loading

The detailed configuration patterns are in modules; load

only the ones you need:

• modules/standard-hooks.md: Layer 1 patterns for

Python, Rust, and TypeScript (load when configuring base

linters).

• modules/component-level-hooks.md: Layer 2 monorepo

scripts and pre-commit wiring (load when project has

multiple components / plugins).

• modules/validation-hooks.md: Layer 3 custom hooks

and SKIP patterns (load when enforcing project conventions

beyond linting).

• modules/ci-integration.md: GitHub Actions workflow

plus a complete .pre-commit-config.yaml example (load

when wiring CI to mirror local checks).

• modules/troubleshooting.md: timing tables, cache

clearing, hook-failure recovery (load when hooks are slow

or failing).

Workflow

1. Create Configuration Files

\\\`bash

Create .pre-commit-config.yaml

python3 plugins/attune/scripts/attune_init.py \\

--lang python \\

--name my-project \\

--path .

Create quality check scripts (for monorepos)

mkdir -p scripts

chmod +x scripts/run-component-*.sh

\\\`

2. Configure Python Type Checking

Create pyproject.toml with strict type checking:

\\\`toml

[tool.mypy]

python_version = "3.12"

warn_return_any = true

warn_unused_configs = true

disallow_untyped_defs = true

strict = true

Per-component configuration

[[tool.mypy.overrides]]

module = "plugins.*"

strict = true

\\\`

3. Configure Testing

\\\`toml

[tool.pytest.ini_options]

testpaths = ["tests"]

pythonpath = ["src"]

addopts = [

"-v", # Verbose output

"--strict-markers", # Strict marker enforcement

"--cov=src", # Coverage for src/

"--cov-report=term", # Terminal coverage report

]

markers = [

"slow: marks tests as slow (deselect with '-m \\"not slow\\"')",

"integration: marks tests as integration tests",

]

\\\`

4. Install and Test Hooks

\\\`bash

Install pre-commit tool

uv sync --extra dev

Install git hooks

uv run pre-commit install

Test on all files (first time)

uv run pre-commit run --all-files

Normal usage - test on staged files

git add .

git commit -m "feat: add feature"

Hooks run automatically

\\\`

5. Create Manual Quality Scripts

For full quality checks (CI/CD, monthly audits):

\\\`bash

#!/bin/bash

scripts/check-all-quality.sh: full quality check for all components

set -e

echo "=== Running Full Quality Checks ==="

./scripts/run-component-lint.sh --all

./scripts/run-component-typecheck.sh --all

./scripts/run-component-tests.sh --all

echo "=== All Quality Checks Passed ==="

\\\`

Hook Execution Order

Pre-commit hooks run in this fixed order; all must pass for

the commit to succeed:

1. File validation (whitespace, EOF, YAML/TOML/JSON syntax)
2. Security scanning (bandit)
3. Global linting (ruff, all files)
4. Global type checking (mypy, all files)
5. Component linting (changed components only)
6. Component type checking (changed components only)
7. Component tests (changed components only)
8. Custom validation (structure, patterns, etc.)

Best Practices

For New Projects

Start with strict settings from the beginning: they are

easier to maintain over time. Configure type checking with

strict = true in pyproject.toml, set up testing early

(include pytest in pre-commit), and document the reason

whenever you must skip a hook.

For Existing Projects

Use a gradual adoption strategy. Start with global checks

(Layer 1), then add component-specific checks (Layer 2)

once legacy issues are resolved. Use --no-verify only for

true emergencies and document why.

For Monorepos and Plugin Architectures

Standardize per-component Makefiles for lint, typecheck,

and test targets. Centralize common settings in a root

pyproject.toml while allowing per-component overrides.

Automate change detection so commits stay fast, and use

progressive disclosure (summary first, detail on failure).

Related Skills

• Skill(attune:project-init): Full project initialization
• Skill(attune:workflow-setup): GitHub Actions setup
• Skill(attune:makefile-generation): Generate component

Makefiles

• Skill(pensive:shell-review): Audit shell scripts for

exit-code and safety issues

See Also

• Quality Gates: three-layer validation: pre-commit

hooks (formatting, linting), CI checks (tests, coverage),

and PR review gates (code quality, security).

Exit Criteria

• [ ] .pre-commit-config.yaml exists at the project root with hooks covering at minimum

Layer 1 (whitespace, YAML/TOML/JSON syntax, global linting).

• [ ] uv run pre-commit run --all-files exits 0 after the configuration is installed,

confirming all hooks pass on the current codebase state.

• [ ] git commit with a staged change triggers the pre-commit hooks automatically (verified

by uv run pre-commit install exit code 0 and presence of .git/hooks/pre-commit).

• [ ] Any --no-verify bypass is absent from the project's documented workflows; if one is

found in existing scripts, it is flagged as a violation rather than silently accepted.