How do I install the Code Better skill?

Run npx skillmds@latest add norravarg/code-better in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Code Better safe to use?

This skill has not completed SkillMD's automated safety review yet. Capability flags: executes scripts. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Code Better?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Code Better free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Code Better?

norraVarg (@norravarg) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Code Better by norravarg — Agent Skill

Name: Code Better
Rating: 4.666666666666667 (3 reviews)
Author: norraVarg

Code Better

Guide the agent to produce code that follows industry best practice and your own customized coding principles. Reviews everything not yet committed — staged, unstaged, and untracked files — against your coding guide, universal correctness/security/testing rules, frontend/backend conventions, language idioms, and your personal rules, verifies each finding independently before trusting it, then helps fix what you approve. Invoke manually after making changes.

norraVarg Updated Sep 22, 2026

File contents

code-better

The guide lives at ${CLAUDE_SKILL_DIR}/guide/. Read ${CLAUDE_SKILL_DIR}/guide/PROCEDURE.md and follow it exactly, step by step.

On this platform, the capabilities PROCEDURE.md step 0 asks you to detect are all available:

Independent check (steps 3 and 4). The Agent tool. In step 3, run the safety pass and the style pass as two separate Agent calls. In step 4, make one separate Agent call per candidate finding, giving it only the file and the quoted rule, not the reasoning that produced the finding. Ask it to try to refute the finding and return exactly one of CONFIRMED, PLAUSIBLE, or REFUTED. Keep confirmed and plausible; drop refuted.
Structured report (step 5). Call the ReportFindings tool once with the surviving findings, ordered most severe first. For each finding: short_summary is the compressed plain-language hook, ending with the severity tag in brackets ("... [High]", "... [Medium]", "... [Low]") — keep the hook itself to roughly 45-50 characters so the tag still fits inside the tool's 60-character limit; summary is one plain sentence stating the defect, with jargon glossed inline on first use, no quoted rule text, and no severity prefix (the tag already lives in short_summary, so it isn't repeated here); failure_scenario is the short, plain-language real-world consequence; category stays the finding-type slug as the tool defines it; verdict stays CONFIRMED/PLAUSIBLE from step 4 — the tool already renders this as a visible badge, which is what keeps plausible findings visibly distinct from confirmed ones. An empty findings list is a valid, complete call — don't skip it or pad it with a manufactured finding.
Diff/preview (step 6). The Edit tool's own diff view.

Input: $ARGUMENTS is the optional extra context PROCEDURE.md's intro paragraph refers to.

Do not summarise guide/PROCEDURE.md or guide/RULE-LAYERS.md from memory. Read them.

---name: code-betterdescription: Guide the agent to produce code that follows industry best practice and your own customized coding principles. Reviews everything not yet committed — staged, unstaged, and untracked files — against your coding guide, universal correctness/security/testing rules, frontend/backend conventions, language idioms, and your personal rules, verifies each finding independently before trusting it, then helps fix what you approve. Invoke manually after making changes.---# code-betterThe guide lives at `${CLAUDE_SKILL_DIR}/guide/`. Read`${CLAUDE_SKILL_DIR}/guide/PROCEDURE.md` and follow it exactly, step by step.On this platform, the capabilities `PROCEDURE.md` step 0 asks you to detect are allavailable:- **Independent check (steps 3 and 4).** The `Agent` tool. In step 3, run the safety pass and the style pass as two separate `Agent` calls. In step 4, make one separate `Agent` call per candidate finding, giving it only the file and the quoted rule, not the reasoning that produced the finding. Ask it to try to refute the finding and return exactly one of `CONFIRMED`, `PLAUSIBLE`, or `REFUTED`. Keep confirmed and plausible; drop refuted.- **Structured report (step 5).** Call the `ReportFindings` tool once with the surviving findings, ordered most severe first. For each finding: `short_summary` is the compressed plain-language hook, ending with the severity tag in brackets ("... [High]", "... [Medium]", "... [Low]") — keep the hook itself to roughly 45-50 characters so the tag still fits inside the tool's 60-character limit; `summary` is one plain sentence stating the defect, with jargon glossed inline on first use, no quoted rule text, and no severity prefix (the tag already lives in `short_summary`, so it isn't repeated here); `failure_scenario` is the short, plain-language real-world consequence; `category` stays the finding-type slug as the tool defines it; `verdict` stays `CONFIRMED`/`PLAUSIBLE` from step 4 — the tool already renders this as a visible badge, which is what keeps plausible findings visibly distinct from confirmed ones. An empty findings list is a valid, complete call — don't skip it or pad it with a manufactured finding.- **Diff/preview (step 6).** The `Edit` tool's own diff view.**Input:** `$ARGUMENTS` is the optional extra context `PROCEDURE.md`'s intro paragraphrefers to.Do not summarise `guide/PROCEDURE.md` or `guide/RULE-LAYERS.md` from memory. Read them.

Frequently asked questions