small, dependency-free tools for people building with claude code. each one does
one job, runs on python3 with nothing to install, and is a single file you can
copy straight into your own setup.
git clone https://github.com/adamentwistle/claude-code-tools
cd claude-code-tools
python3 slop-detector/slop_detector.py --help
python 3.9+, no dependencies, no network. every tool also works on its own if you just copy the one script.
the first four are a doc-quality suite: three linters and one command that runs all of them as a single gate.
| tool | what it does |
|---|---|
| slop-detector | scores any text for the fingerprints of machine-written prose and gives a weighted, line-numbered report with a ci-gateable exit code. the shared engine the others build on. |
| claude-md-linter | scores a CLAUDE.md out of 100 for what actually makes one good: tight, concrete, scannable. flags bloat and vague directives by line. |
| readme-hero-lint | scores a readme's hero (everything above the first scroll) for first impression: does it say what the thing is, show how to run it, and skip the buried lede. |
| lint-all-docs | one command that composes the three linters above into a single pass/fail over your repo's docs, for ci or a pre-commit hook. |
| mcp-readonly-skeleton | a dependency-free mcp stdio server that exposes read-only tools and nothing else. the safe shape to copy from when you start a discovery-style server. |
lint-all-docs lints this repo's own docs. one check fails on purpose: the
slop-detector readme spells out the very patterns it scans for, so it trips its
own scanner. that is the tool working as intended. skip it for a clean pass:
python3 lint-all-docs/lintall.py --exclude slop-detector/README.md
the agent-engineering side lives in a sibling repo: agent-eng-kit, stdlib-only tools for eval, observability and safety of agentic systems.
built in public by @aae_on_x: an engineer building agentic systems with claude code. this is where the small pieces land as they ship, given away one at a time.