You type “fix the login thing.”
It knows which one.

Sharpen turns the prompt you typed into the prompt you meant, and shows it to you before any work starts.

/plugin marketplace add melthond/sharpen /plugin install sharpen@sharpen

01 How it works

One gate before every task.

Sharpen reads your request and takes one of three paths: keep it, fill it in, or ask. Pick a request to see the card it would show.

Ambiguous · asksIllustration
Which did you mean?
  • The PIN lockout
  • The expired member session
  • Run as typed
  1. 01

    Gate

    Clear requests keep their wording. One reading gets the gaps filled. Two readings get a question.

  2. 02

    Gather

    The conversation first, then a few focused code searches, then your skills, tools and hooks.

  3. 03

    Engineer

    Target, files, limits, tools, a finish line and evidence, checked for common prompt problems.

  4. 04

    Review

    The prompt lands on a card. Nothing starts until you send it, edit it, or run your original.

02 What it adds

Same request. A prompt that can finish.

A real run from the Sharpen README: one line in, a brief with the file, the related work, the rule to keep and the check out.

You type

“fix the clipboard so it clears after 30 seconds”

Sharpen hands Claude
  • Change the clipboard clear for copied keys from 60 to 30 seconds.
  • Where: apps/admin/src/lib/clipboard.ts:2 (HOLD_MS = 60_000 → 30_000)
  • Also update: generate.tsx:181, clipboard.test.ts, PRD FR-19, checklist US-020.
  • Keep: the "leave it alone if something else was copied" rule.
  • Hooks: dead-code-check runs after every edit.
  • Check: the clipboard test passes.
  • Knows which one

    Two open login bugs? It asks which you meant, with one option per reading.

  • Full briefs

    Starting state, target state, scope lock, stop-and-ask rules and evidence for anything that edits code.

  • No guessed values

    Prices, limits and wording only you can decide become visible [ASK: …] blanks.

  • 88 anti-patterns

    Vague verbs, missing finish lines, relative dates, test gaming and third-try corrections are caught before you see the card.

  • Techniques built in

    Escalation, gated chains, search budgets, context limits and long-run plans, drawn from the major prompting guides.

  • Your setup, used

    Tools: and Hooks: lines name the skills, agents and hooks you actually have installed.

  • Knows the phase

    Kickoff, spec, plan, tasks or build: each gets its own shape and a next step. Small fixes stay small.

  • Safety checks

    Risky verbs get confirmed, new packages get checked in the registry, and private data plus outward sends get a stop.

  • Grounded in research

    Built from the Anthropic, OpenAI and Google guides, published studies and 2025–2026 developer surveys, each linked in the references.

03 Companion skills

Ten skills, one plugin.
Use one, or all of them.

Sharpen runs on its own. These run when you call them: type /sharpen: in Claude Code, whether automatic review is on or off.

sharpen-prompt

Turn a rough request into a standalone brief for another AI, a ticket, or a teammate.

Best forWrite a prompt for another reader

/sharpen:sharpen-prompt fix the empty state — for a ticket
Read the docs

You: /sharpen:sharpen-prompt fix the empty state — for a ticket

Sharpen

Prompt for a ticket
Goal
Show an empty state when the list has no items.
Context
What a reader outside this session needs to know.
Done when
The empty state renders with a next step.

Example. Handed back as text. Nothing runs.

sharpen-decompile

Break an existing prompt into its parts, name the weak spots, and adapt, simplify, or split it.

Best forUnderstand or adapt a prompt

/sharpen:sharpen-decompile make the dashboard better
Read the docs

You: /sharpen:sharpen-decompile make the dashboard better

Sharpen

Prompt breakdown
Parts
Verb “make better”, target “dashboard”.
Weak spot
“Better” has no measure.
Weak spot
No finish line.

Example. The prompt is analyzed, never run.

sharpen-handoff

Carry decisions, work tried, failures, and the next move into one prompt for a new session.

Best forMove work into a fresh session

/sharpen:sharpen-handoff
Read the docs

You: /sharpen:sharpen-handoff

Sharpen

Handoff prompt
Decided
Choices already made, so they stay made.
Tried
What failed, so it isn’t retried.
Next
The next step and its check.

Example. Paste it into a fresh session or hand it to a teammate.

sharpen-split

Turn a large request into ordered, bounded prompts grouped into waves that can run in parallel.

Best forBreak a large request into waves

/sharpen:sharpen-split rebuild settings, tests, and docs
Read the docs

You: /sharpen:sharpen-split rebuild settings, tests, and docs

Sharpen

Split into waves
Wave 1
Rebuild settings.
Wave 2
Tests and docs, in parallel.
Each
Its own finish line and scope.

Example. Sharpen writes the plan; you run the prompts.

sharpen-interview

Answer one focused question at a time, then get a clear feature specification with decisions captured.

Best forTurn a fuzzy feature into a spec

/sharpen:sharpen-interview add team invitations
Read the docs

You: /sharpen:sharpen-interview add team invitations

Sharpen

Question 1
Asks
Who can send an invitation?
Then
One question at a time, only the ones that matter.
Writes
SPEC.md once the decisions are clear.

Example. It never builds the feature during the interview.

sharpen-plan

Get the kickoff rules, spec, plan, tasks and build prompt a project needs, sized to the work and written only after you approve.

Best forPlan a project in phases

/sharpen:sharpen-plan team invitations with roles
Read the docs

You: /sharpen:sharpen-plan team invitations with roles

Sharpen

Project plan
Size
Big feature: the full set of files.
Files
Kickoff rules, spec, plan, tasks, build prompt.
Shown
Every file before it is written.

Example. Nothing is written until you approve.

sharpen-review

Review recent prompts, see what was missing, and get one practical improvement to try next.

Best forImprove how you prompt

/sharpen:sharpen-review
Read the docs

You: /sharpen:sharpen-review

Sharpen

Prompt review
Missing
What recent prompts left out.
Cost
What that cost you.
Habit
One practical thing to try next.

Example. Coaching on this conversation’s prompts.

sharpen-save

Package a prompt you use often as a project slash command, with a preview before it is written.

Best forSave a prompt you repeat

/sharpen:sharpen-save check-release
Read the docs

You: /sharpen:sharpen-save check-release

Sharpen

New slash command
Makes
A project slash command.
Generalizes
The details that change each time.
Shows
The proposed file first.

Example. It asks before writing anything.

sharpen-audit

Surface rules, checks, and project knowledge an agent cannot infer, with paste-ready CLAUDE.md notes.

Best forFind project knowledge an agent cannot see

/sharpen:sharpen-audit
Read the docs

You: /sharpen:sharpen-audit

Sharpen

Setup audit
Checks
Guidance, test commands, conventions, tools.
Ranks
What to add first.
Gives
Paste-ready CLAUDE.md notes.

Example. The audit changes no files.

sharpen-help

Get a compact reference for Sharpen controls, companion skills, and how to use each one.

Best forGet the quick reference

/sharpen:sharpen-help
Read the docs

You: /sharpen:sharpen-help

Sharpen

Sharpen help
State
Whether automatic review is on.
Controls
How to turn it on, off, or check it.
Skills
Every companion skill, in one place.

Example. Everything in one compact reference.

04 FAQ

Questions, answered.

For a clear request, it keeps the ask intact and adds helpful context. If a rewrite could change what you meant, you can choose the original from the card.

There is one review decision per sharpened prompt. Use raw: for a request you want to send unchanged, or turn automatic review off with /sharpen off.

It uses some of your current agent session's context: the method and setup summary load once per session, with a short reminder on each prompt. There is no separate Sharpen API or service fee.

It uses what Claude Code lists for skills, agents, MCP tools, and plugin commands, plus the hook and enabled-plugin summary it reads at session start. It does not execute hooks while gathering context.

The current plugin is for Claude Code. The project describes support for other agents as future work; do not assume the automatic review flow is available elsewhere today.

From the Anthropic, OpenAI and Google prompting guides, open-source projects such as the DAIR.AI Prompt Engineering Guide, the OpenAI Cookbook and GitHub Spec Kit, published studies on prompt and agent failures, and 2025–2026 developer surveys. CREDITS.md lists every source and what it contributed, and model-specific settings carry the date they were checked.

No. Nothing leaves your machine. /sharpen shows two local counts, the review cards shown and the follow-ups that pushed back, so you can see whether sharpened prompts land the first time. No prompt text is stored.

No prompt tool can guarantee that. Sharpen makes the requested work, boundaries, and checks clearer, and lets you review the prompt before it starts. The agent still needs to do and report the work.