one-click-improve

fela@agents:~$ oci --surfaces vscode,cursor,claude,codex

VS Code / Cursor / Claude Code / Codex

One-Click Improve

Turn rough coding prompts into context-aware engineering tasks before they reach your agent. It routes through a harness intelligence library, loads embedded skills, asks for missing details, and ships a prompt with acceptance criteria and validation steps.

install compare harness
install now 
curl -fsSL https://improv.sh/install.sh | bash

outcome demo

A vague prompt burns turns. A prepared prompt gets the agent working.

Two examples show the difference between sending a raw request straight to an agent and sending the task after One-Click Improve has shaped it for the active harness.

rough prompt Fix login bug

Direct to agent

First turn: clarify or guess The agent still has to discover scope before useful work can start. 
Agent first response:
Which login flow is broken?
What error do you see?
Which files should I inspect?

Likely outcome:
2-4 messages spent finding the task.
weak spots caught
  • Missing affected flow
  • No error state
  • No done condition
  • No validation path

Through One-Click Improve

First turn: execute with guardrails The agent receives a scoped task, acceptance criteria, and validation plan. 
One-Click output:
Objective: fix the failing auth path.
Context to confirm: flow, error, auth files.
Guardrails: preserve sessions; avoid unrelated UI changes.
Done when: login works and the failure has a targeted test.
what gets added
  • Explicit scope
  • Harness-ready wording
  • Acceptance criteria
  • Validation steps

prompt-prep loop

From vague request to runnable task

01

$ capture

Select text, type `@im`, run `/im`, or call `@im` in the terminal.

02

$ analyze

Score clarity, specificity, context, constraints, success criteria, testing, ambiguity, and architecture.

03

$ load skills

Use the bundled improve-prompt skill plus the harness route: VS Code, Cursor, Claude Code, Codex, or shell.

04

$ apply

Copy, apply to the editor, or send the improved prompt to the agent surface you already use.

harness intelligence library

It carries context, skills, and harness rules into the prompt.

One-Click Improve knows where the prompt is going: VS Code, Cursor, Claude Code, Codex, or a terminal flow. It uses the right route for that surface, applies the embedded prompt-improvement skill, and turns the request into something the receiving agent can act on.

@im improve the prompt @harness choose the best route /im inside Cursor or Claude Code
context Selection, original request, missing details, current workflow, and assumptions that must be made explicit.
harness VS Code `@im` and `@harness`, Cursor `/im` and `/harness`, terminal `@im` and `@harness`, Claude Code commands, Codex-ready output.
skills Bundled `improve-prompt` skill synced across `.claude`, `.cursor`, `.agents`, and `.github`.
guardrails Offline local provider by default, optional model adapters, SecretStorage keys, no silent prompt sending.

works where prompts start

One skill, many harnesses

VS Code

Use the sidebar panel, improve selected text, or mention `@im` in VS Code Chat.

@im Fix login bug

Cursor

Install the VSIX and use the Cursor command route for reliable chat prompting.

/im Add tests for session timeout

Terminal

Generate an improved prompt from any shell and paste or pipe it into Claude Code or Codex.

@im "Refactor dashboard loading"

Harness Skills

Ask `@harness` which route to use, then sync accepted skills into `.claude`, `.cursor`, `.agents`, and `.github`.

@harness "I am working in Cursor"

local first

Offline by default, model-backed when you choose

The deterministic local provider ships ready to use. OpenAI and Anthropic adapters stay dormant until you opt in and store keys with VS Code SecretStorage.

localNo network, deterministic output
openaiOptional skill-guided model rewrite
anthropicOptional Claude API rewrite

start here

Install once, improve prompts everywhere

The installer detects VS Code and Cursor, installs the extension, creates terminal shortcuts, and syncs the bundled skill to user-level harness folders. 

curl -fsSL https://improv.sh/install.sh | bash