prompt-optimizer

Transforms rough requests into optimized, well-structured prompts for Claude CoWork and Chat. Conducts an adaptive interview to fill gaps, invokes the right skills automatically, and delivers a polished prompt ready to run.

Version: 0.1.0
Author: MessagePoint Media
Repo: https://git.alwisp.com/jason/prompt-optimizer
CoWork Project: CW-006 — Prompt Optimizer

Overview

The Prompt Optimizer is a skill-only CoWork plugin for the MPM team. It solves the problem of vague or underspecified prompts — when a team member knows what they want but isn't sure how to ask Claude for it effectively. The skill conducts an adaptive interview (calibrated by experience level: Beginner, Intermediate, or Advanced), automatically identifies which installed MPM skills should be invoked for the task, rewrites the rough request as a polished and complete prompt, shows it for review before running anything, and saves the optimized prompt as a reusable .md file. It integrates with all core MPM skills including docx, pptx, xlsx, pdf, mpm-brand-voice, odoo-mpm, and display-catalog. No credentials or MCP server are required — this is a skill-only plugin.

Skills

Skill What It Does

optimize-prompt Full prompt optimization workflow. Triggers on: "optimize this prompt", "improve my request", "help me ask this better", "make this clearer", "polish my prompt", "help me write a better question", "prompt optimizer", "I'm not getting what I want from Claude", or any rough idea the user wants refined into a well-structured request. Conducts an adaptive interview (up to 5 questions for Beginners, 3 for Intermediate, 2 for Advanced), silently analyzes the raw prompt for gaps in clarity/audience/purpose/constraints/format, identifies applicable skills, rewrites the prompt with full 6-part structure, presents original vs. optimized for review with a bullet list of improvements, saves as .md file, and executes on explicit approval.

Skill	What It Does
`optimize-prompt`	Full prompt optimization workflow. Triggers on: "optimize this prompt", "improve my request", "help me ask this better", "make this clearer", "polish my prompt", "help me write a better question", "prompt optimizer", "I'm not getting what I want from Claude", or any rough idea the user wants refined into a well-structured request. Conducts an adaptive interview (up to 5 questions for Beginners, 3 for Intermediate, 2 for Advanced), silently analyzes the raw prompt for gaps in clarity/audience/purpose/constraints/format, identifies applicable skills, rewrites the prompt with full 6-part structure, presents original vs. optimized for review with a bullet list of improvements, saves as `.md` file, and executes on explicit approval.

Workflow

Expertise calibration — Asks the user their experience level (Beginner / Intermediate / Advanced). Controls how many questions are asked and how much explanation is given.
Silent gap analysis — Evaluates the raw prompt internally against: clarity, audience, purpose/context, constraints, output format, and applicable skills. Ranks missing items by importance.
Adaptive interview — Asks one question at a time, stopping when enough information is gathered or the maximum question count is reached. Never asks about things already specified.
Skill identification — Consults references/skill-awareness.md to determine which installed skills to invoke. Trigger phrases are woven naturally into the optimized prompt — users never need to know skill names.
Prompt rewrite — Produces a complete prompt with: context/role setup, core request, audience & purpose, constraints, output specification, and skill invocations.
Review presentation — Shows original vs. optimized side-by-side with a bullet list of specific improvements. Beginners get an explanation of the most impactful change.
Save as .md — Simultaneously saves to optimized-prompt-[topic]-[date].md in the outputs folder with a download link.
Execute on approval — Runs the optimized prompt directly when the user approves ("run it", "looks good", "yes", etc.) with no preamble.

Skills Integrated

The optimizer is aware of all MPM-installed skills and includes the right ones automatically:

Skill	Trigger Condition
`docx`	User wants a Word document, report, memo, letter, proposal, or any `.docx` file
`pptx`	User wants a PowerPoint, slide deck, pitch deck, or any `.pptx` file
`xlsx`	User wants an Excel file, spreadsheet, budget, tracker, or any `.xlsx` file
`pdf`	User wants a PDF, or to merge/split/extract from PDFs
`mpm-brand-voice:brand-voice-enforcement`	Any external-facing content, client communications, or marketing materials
`odoo-mpm:odoo`	Prompts involving MPM business data (contacts, projects, CRM, inventory, orders)
`display-catalog`	Prompts requiring MPM display product specs (never routes to web search for MPM products)

Reference Files

File	Purpose
`skills/optimize-prompt/references/skill-awareness.md`	Maps all installed MPM skills to trigger conditions; includes critical rule that MPM product specs must come from internal sources only, never web search
`skills/optimize-prompt/references/optimization-guide.md`	Best-practice prompt structure, language patterns, anti-patterns to avoid, before/after examples, tone calibration guide, output format specification reference

Setup Instructions

Upload the .plugin file via the Plugins panel in Claude CoWork
Restart CoWork to activate the skill
Verify activation: say "prompt optimizer" — the skill should respond asking for your rough prompt
No credentials, environment variables, or MCP server configuration required

Setup Checklist

.plugin file uploaded in CoWork Plugins panel
CoWork restarted after installation
Skill verified: say "prompt optimizer" and confirm it responds

Connection Details

Field	Value
Plugin type	Skill-only (no MCP server)
Auth required	None
External APIs	None
Output artifacts	Saves `.md` files to the CoWork outputs folder
MCP server name	N/A

Workflow Notes

MPM product specs: The optimizer will never direct Claude to search the web for MPM product information. All product spec requests are routed to the Display Catalog skill (preferred) or Odoo product catalog as fallback.
Skill triggers are natural language: Trigger phrases are woven into the optimized prompt naturally — users never see or need to know skill names.
Experience level resets each session: The expertise calibration question is asked at the start of each optimization; it does not persist.
Always review before execution: The optimized prompt is shown for review every time. Execution only happens on explicit user approval.
Adjustments without re-interview: If the user requests changes to the optimized prompt, the skill incorporates feedback and re-presents without re-running the full interview.

Requirements

Claude CoWork (desktop app)
No additional dependencies or packages

Troubleshooting

Symptom	Fix
Skill doesn't trigger on "prompt optimizer"	Verify `.plugin` file is installed and CoWork has been restarted
Skill triggers but doesn't invoke other skills (e.g., docx, pptx)	Confirm those skills are also installed separately; prompt optimizer requires them to be present
Optimized prompt saves but doesn't execute	Say "run it" or "looks good" — the skill waits for explicit approval before executing
Skill asks too many questions	Check expertise level selection — Advanced mode asks only 1–2 questions

6.9 KiB Raw Blame History Unescape Escape