GitHub

Guidance, skills, and scripts that help coding agents (Claude Code, OpenCode, or any other) deliver consistently high-quality code — designed so that even small-context models can complete features and fixes to a high standard.

What's here

Path What it is
agents-files/ Entry-point instructions to copy/merge into your project (local/ for small-context agents, remote/ for capable ones, plus a Claude Code variant)
guidance/ Condensed standards and process docs, one topic per file, routed by guidance/index.md
skills/ Step-by-step playbooks for multi-step tasks (project setup, CI, reviews, deploys, ADRs, ...), routed by skills/index.md
scripts/ The automation: check.sh (quality gate), run-tests.sh, check-prereqs.sh, check-install.sh, setup-hooks.sh, git hooks
adapters/ Wiring for specific agent platforms (Claude Code skills + hooks, OpenCode)
configs/ Shared tool config used by the gate when a project supplies none (currently markdownlint)

Install into a project

git submodule add https://github.com/instantiator/dev-qual
./dev-qual/install.sh

The installer is interactive: it asks which agent tier(s) and platforms you use, merges entry files into your repo (never clobbering existing AGENTS.md/CLAUDE.md), and offers to install the git hooks that run the quality gate on commit and push. Re-run it any time.

Prefer manual setup? Copy a file from agents-files/ to your repo root, then run ./dev-qual/scripts/setup-hooks.sh.

Keep it up to date

git submodule update --remote dev-qual && ./dev-qual/install.sh

Installed before the rename, as dev-environment? Move the submodule and re-run the installer — it replaces the old marker block rather than duplicating it:

git mv dev-environment dev-qual
git submodule set-url dev-qual https://github.com/instantiator/dev-qual.git
./dev-qual/install.sh

scripts/check-install.sh reports what in your project has drifted from the checkout — an edited CLAUDE.md, copied hooks, a skill that is no longer a symlink — so you can merge rather than overwrite. The update-latest skill walks an agent through it.

The quality gate

scripts/check.sh detects every stack in your project — Node/TypeScript, .NET, Python, plus shell and markdown wherever they appear — and runs format check → lint → typecheck → build → unit tests → aislop scan, printing PASS/FAIL with a fix-hint per failure. Missing tools SKIP with an install command rather than breaking the run.

Mode Adds Run by
--fast format, lint, typecheck pre-commit hook, after each unit of work
(default) build, unit tests, aislop scan pre-push hook
--comprehensive every test suite, package security audit you, after the last unit of work

The git hooks run the first two automatically, so quality doesn't depend on anyone, human or model, remembering. pre-commit also runs scripts/pre-commit-fixups.sh if your project has one — formatting, generated files, licence lists — and re-stages what it changed, leaving partially-staged files alone.

Third party tools

Some of these rules refer to and lean on third party tools. With gratitude:

Tool Description License
aislop Catch the slop AI coding agents leave in your code: narrative comments, swallowed exceptions, as-any casts, dead code, oversized functions. 50+ rules across 8 languages. MIT
ponytail Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote. MIT

Read the original on github.com ↗