Claude Code Plugin

PromptOps

Evaluate, improve, and run your prompts — all in one session. No setup required.

$ gh repo clone harmansidhudev/promptops copy
The Loop

Four commands. One session.

/evaluate
→
/improve
→
/regression
→
/run
Commands

What you get

/evaluate
Score any prompt 1–5 across five dimensions. See the top 3 issues with copy-paste fixes.
/improve
Paste a rough prompt, get a polished version with schemas, guardrails, and zero filler.
/regression
Compare old vs new. See what improved, what regressed, and whether it's safe to ship.
/compare
Benchmark across Opus, Sonnet, Haiku, GPT-4o. Per-run cost, daily cost, routing rules.
/run
Execute the winning prompt for real. Delivers actual output, not another evaluation.
In Action

See the loop

Claude Code — promptops
> /evaluate You are a code reviewer. Find bugs.

Prompt Evaluation — 1.8/5.0
Structure 2/5 · Specificity 2/5 · Output Control 1/5
Error Prevention 1/5 · Testability 2/5

1. [Output Control] No format specified. Fix: add JSON schema
2. [Error Prevention] No guardrails. Fix: add "Do NOT" patterns
3. [Specificity] Too vague. Fix: add role framing + constraints

> /improve You are a code reviewer. Find bugs.

Improved Prompt — 1.8 → 4.2
+Structure · +Output Control · +Error Prevention · −Filler

> /run
Running the improved prompt now...
Auto-Invoked

Skills that work in the background

auto-invoked

Prompt Quality Check

Spots weak prompts while you write them. One-line fix when it matters. Silent when your prompt is solid.

auto-invoked

Golden Dataset Builder

Saves approved outputs as test cases. Builds evaluation data over time with zero extra effort.

Try it now

Open source. No setup. Works in Claude Code terminal and Mac app.

View on GitHub Installation Guide