Prompt improver

Skill shinzoxD/knackbox/skills/productivity/prompt-improver

Curated, auditable, benchmark-ready Agent Skills library for Claude Code, Codex, OpenCode, Cursor, and more.

Install

npx -y skills add shinzoxD/knackbox --skill prompt-improver

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

28 days oldThe repository was created 28 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Rewrite and strengthen prompts for AI models — diagnose what's vague, add structure, and return a copy-paste-ready improved prompt. Use whenever the user shares a prompt to improve, asks why an AI's output was bad or inconsistent, wants a system prompt written, or says "make this prompt better" for any model or agent.

The file declares its own license as Apache-2.0. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

2.7 KB, 492 tokens by cl100k_base, as published. Nobody here has run it

Prompt Improver

Most bad outputs are underspecified inputs. Your job: find what the prompt leaves the model to guess, then remove the guessing.

Process

Diagnose first. Against the user's stated goal, check the prompt for: missing context, undefined audience, no output format, no success criteria, no examples, buried or conflicting instructions, and scope so broad the model must gamble.
Rewrite using this skeleton (drop parts that don't apply):
- Role/expertise the model should adopt
- Context: background the model can't infer
- Task: one unambiguous instruction, imperative voice
- Constraints: length, tone, what to avoid, edge-case handling
- Output format: exact structure, ideally with a mini example
- Quality bar: what distinguishes a great answer from an okay one
Explain the changes in 3–5 bullets: what was ambiguous → what you pinned down. This teaches the user to self-serve next time.

Rules

Preserve the user's intent exactly; improving a prompt never means changing what it's for. When intent itself is unclear, ask one question rather than guessing.
One or two well-chosen examples ("few-shot") beat three paragraphs of description — include them when format or style matters.
For tasks with reasoning, instruct the model to work step-by-step before giving the final answer, and to separate the two.
Kill instruction conflicts ("be comprehensive but keep it to one line") — surface the tension and pick a side with the user's goal.
If the original prompt is already good, say so and offer at most two marginal upgrades. Don't rewrite for the sake of it.
Deliver the improved prompt in a single copy-pasteable block, ready to use with no editing except clearly marked [placeholders].

Edge cases

"It works sometimes": inconsistency usually means missing format spec or examples — fix those first.
Prompt for an agent/tool-using model: add when-to-use-which-tool guidance and stop conditions.
User wants shorter, not better: compress by cutting redundancy, never by removing constraints that carry behavior.