AIM+ Prompt Builder

What this skill does

• Builds prompts from rough ideas for ChatGPT, Claude, Claude Code, Codex, Gemini, Cursor, Copilot, Windsurf or any other AI
• Freshness check on every request, research in official sources only for time-sensitive topics, dated and condensed
• At most 1 to 3 targeted questions with a recommended option, native question buttons where available, often none
• Improves, repairs, shortens, audits and adapts prompts between tools, with a short change list
• Removes prompt myths: hype personas, tipping, all-caps commands, forced step-by-step reasoning
• Plans in real tools such as skills, MCP, web search or structured outputs, but only ones that exist
• 12-point quality gate plus the prompt-lint.mjs script with 39 automated tests
• You steer in plain words: "no research", "don't ask", "just the prompt", "two variants", "for Claude Code"

Description

AIM+ Prompt Builder is an AI prompt builder and prompt improver that learns what matters before it writes. You don't need to study prompt engineering. Just say "write me a prompt for a new landing page", "I want to build a web app with Codex" or "here's my prompt, make it better", and the skill works out the real task, what is already known and what is missing. Its starting point is the simple Actor, Input, Mission structure: which perspective helps, what to work with, what exactly to produce. The "+" adds what modern AI work really needs: context, audience, constraints, output format, success criteria, available tools and safety. It only adds what improves the task, because a short, clear prompt beats a long template. Every request gets a freshness check. Rewriting a paragraph needs no research, but for APIs, frameworks, AI tools, SEO and GEO, social platforms, regulation or pricing it runs short research in official sources and keeps only the findings that change the prompt, dated and condensed. If your AI has no web search, it says so openly and puts a verification step into the prompt instead. It never claims a check that didn't happen. When something important is missing, it asks at most one to three targeted questions with a recommended answer, ranked by how much they change the result and specific to your task rather than a generic questionnaire. It uses your AI's built-in question buttons where they exist and a numbered list elsewhere, and it asks nothing when your request is already complete. Existing prompts get improved, repaired, shortened or audited, or adapted to another tool, for example from ChatGPT to Claude Code. It changes only what needs changing and strips prompt myths with no evidence behind them, such as "you are the world's best expert", tipping, all-caps commands or forced step-by-step thinking. Every prompt passes a 12-point quality gate from goal clarity to non-contradiction, backed by the prompt-lint.mjs script. You get a copy-ready prompt with a short rationale, or truly just the prompt if you ask.

Examples

Tested with Claude Code

Write a LinkedIn post announcing our new CRM feature: [name, what it does, who it's for].

Audience: [role, company size] who use a CRM every day.
Goal: get them to [try the feature / book a demo].
Length: 120 to 180 words, short paragraphs, exactly one clear call to action at the end, at most 3 hashtags.
Tone: confident and concrete. Use one real benefit or number from the facts above and invent none.

Illustrative answers. Results may differ depending on the AI and model.

What it does not do

• No faked research: no "verified" without a real check
• No perfect-prompt promise: better, more complete prompts; results still depend on the model and your input
• No jailbreaks: no "god mode", no policy bypass, no extraction of hidden reasoning
• No image prompts: that's what Image Prompt Director is for
• No saved settings: preferences last for the session; for permanent ones you get a profile block to paste

Compatibility & tech

Tested (internal)

6 scenarios

Recommended runtime

Works with any current model (Claude, GPT, Gemini). An AI with web search gets the most out of the freshness check. The lint script needs Node without extra packages and is optional.

Modes

Create from a rough idea · Improve and repair · Expand or simplify · Adapt to another model or tool · Refresh with current facts · Audit (findings only)

Inputs

a rough idea or request in plain language · an existing prompt or system prompt · files, code or project context to build on

Output format

A copy-ready prompt in its own block, followed by visible assumptions and at most five short points on why it is stronger. An optional one-line freshness status. In audit mode a findings table ranked by impact; on request only the prompt.

Subcategory

Prompt generator, prompt optimization and context engineering

License

Proprietär

Security profile

What you get

• aim-prompt-builder-1.0.0/13 files
  • .claude-plugin/marketplace.json
  • plugins/aim-prompt-builder/skills/aim-prompt-builder/13 files

    • SKILL.md
    • manifest.json
    • data/3 files

      • tests/3 files

        • benchmark-results.md
        • benchmark.json
        • lint-cases.json
    • references/7 files

      • aim-plus-model.md
      • evidence-and-myths.md
      • freshness-research.md
      • prompt-architecture.md
      • questions.md
      • security-privacy.md
      • targets.md
    • scripts/1 file

      • prompt-lint.mjs
  • .agents/skills/aim-prompt-builder/→ universal, same content (Codex, Cursor, Copilot, Gemini, Windsurf, Cline)
  • LICENSE.txt

Installation

Also works as a chat prompt

Note

Actor, Input and Mission is a common structure, not an official standard of any AI provider. Current research is only possible if your AI has web search; without it, the skill adds a verification step to the prompt. The skill is not legal, medical or financial advice, even when the prompt covers those topics.

Changelog

• v1.0.023.09.2026Initial release: turns a rough idea or an existing prompt into a clear, current, ready-to-run prompt for any AI. Built on Actor, Input and Mission plus only the sections that improve the task. Freshness check on every request, short primary-source research only for time-sensitive topics, never faked research. At most 1 to 3 targeted questions via the platform's native question UI or a numbered list, often none. Modes to create, improve, repair, shorten, adapt to another tool and audit. Removes prompt myths. 12-point quality gate, prompt-lint.mjs with 39 tests, 7 references with dated evidence and platform notes.

Frequently asked questions