Skip to content
Conduit

skill

kiss-skill

KISS (Keep It Simple, Stupid) skill for Claude Code, Codex CLI and Hermes Agent: checks whether a feature, layer or dependency is needed before building it.

jenluvsem/kiss-skillcommunitylistedunreviewed
Use skill

Set up

Skills are installed by placing their folder where your agent loads skills.

Download ZIP from GitHub Open the repository

The download comes from github.com/jenluvsem/kiss-skill itself, not from our server, so what you get is exactly what its author published.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below.

README

synced Oct 28, 2026

KISS

The name: Keep It Simple, Stupid. Kelly Johnson's line, from Lockheed's Skunk Works (U-2, SR-71). He handed his engineers a few tools and said a mechanic in the field fixes this jet with these, nothing else. "Stupid" was about that toolbox. Not the mechanic. Story's on Wikipedia.

This repo is that rule as a skill. Claude Code, OpenAI Codex CLI, Hermes Agent. Before the agent builds a feature, a layer, a config file or a dependency, it asks: who needs this today? Nobody? Then you get the smallest change that fixes today's problem. Plus the point at which the bigger version starts paying off.

Same file for all three: skills/kiss/SKILL.md. Id is lowercase kiss; the Agent Skills spec forbids capitals.

When it kicks in

On its own when you ask for a plugin system, config file, abstraction, layer or dependency. When you weigh two designs. When you want a plan or diff reviewed. Or call it by name, see Use.

What it makes the agent do

Ladder. Top down, stop at the first rung that covers the whole job:

  1. Skip it. One line on what got skipped.
  2. Reuse what the project has.
  3. Stdlib or platform.
  4. A dependency already installed.
  5. Least new code that works.
  6. Run it on real input. Read the output.

Then 14 rules, each ending in a yes/no check. Two of them: "Can you point at someone who needs it today?" "Would the shared function need a flag per caller?"

Then 8 smells. Interface with one implementation. Wrapper that just forwards a call. "We might need it for...". Five more in SKILL.md.

Never cut: validation at trust boundaries, error handling that guards data, security, accessibility, anything you explicitly asked for.

You get back a plan, why, when to build more, what was skipped. Still want the big version? The agent names the cost, then builds it.

A real run

Fixture: report.py, 16 lines, prints a CSV. Prompt:

Add a plugin system with YAML config to this script so it can support more output formats later.

Claude Code, Sonnet, no skill: format registry, YAML config, plugins loaded via importlib. PyYAML added.

Same, with KISS: a FORMATS = {"text": text} dict and a --format flag. Nothing added. Plugins wait until someone outside the repo must add formats without editing the script. Quote from that run: "The script is 15 lines and has one output format. No one has asked for a second format, so a plugin loader and a config file would be bigger than the script and have nothing to configure."

Tested in all three, same fixture:

  • Claude Code 2.1.304 on Sonnet found the skill and loaded it in 6 of 6 runs. No plugins, no YAML.
  • Hermes Agent 0.22.2 on Sonnet: 5 of 5 runs. No plugins, no YAML, about 10 lines changed.
  • Codex CLI 0.163.0 on MiniMax M3.1 Flash (free, through OpenRouter): one run. It dropped the plugins but kept YAML and a registry.

Codex was weakest. One run, small free model. A floor, not a verdict.

Install

Claude Code

Plugin:

/plugin marketplace add jenluvsem/kiss-skill
/plugin install kiss@kiss-skill

Personal skill:

git clone https://github.com/jenluvsem/kiss-skill
cp -r kiss-skill/skills/kiss ~/.claude/skills/kiss

/kiss:kiss for the plugin, /kiss for the personal one.

OpenAI Codex CLI

git clone https://github.com/jenluvsem/kiss-skill
mkdir -p ~/.agents/skills
cp -r kiss-skill/skills/kiss ~/.agents/skills/kiss

One repo only? .agents/skills/kiss inside it. No $kiss? Restart Codex, or use /skills.

Hermes Agent

hermes skills install jenluvsem/kiss-skill/skills/kiss

Scanned before install. Tap also works: hermes skills tap add jenluvsem/kiss-skill. Got ~/.agents/skills in skills.external_dirs? Then Hermes reads the Codex copy. One install, two agents.

Use

Mostly automatic. By hand:

Use the kiss skill to review this plan before I start.

$kiss should this be a new service or a function in the existing one?

Repo

skills/kiss/SKILL.md              the skill, ~110 lines, all three tools
skills/kiss/references/bible.md   long form: 18 rules, sourced
.claude-plugin/                   Claude Code plugin + marketplace
docs/compatibility.md             per-tool format notes, official links
tests/validate_skill.py           stdlib validator

bible.md only gets opened when a rule needs its source. I checked each quote on the original page.

Validate

python3 tests/validate_skill.py
python3 -m unittest discover -s tests

CI runs both, every push.

Sources

Kelly Johnson, Dijkstra, Hoare, Eric Raymond, Martin Fowler, Kent Beck, John Gall, Brian Kernighan, Rich Hickey, Fred Brooks, Google SRE book, John Ousterhout, Sandi Metz, Dan McKinley, Rob Pike, Donald Knuth, PEP 20, Richard Gabriel. Links: skills/kiss/references/bible.md.

License

MIT. LICENSE.

↑↓ navigate openesc close