Promptly

A Claude Code skill that turns messy exploratory conversations into reusable single-shot prompts.

The Problem

When you explore a problem with Claude — iterating, clarifying, refining — you often end up with a great output. But next time you need the same thing, you start from scratch.

Promptly captures those patterns.

How It Works

After a successful exploration, run /promptly and it will:

1. Analyze the conversation journey
2. Distill a refined, reusable prompt
3. Save it to your prompt library with trigger keywords

Next time you start a similar task, it offers your refined version.

Features

Mode	What It Does
Distill (default)	Extract reusable prompts from exploratory conversations
Match	Surface relevant prompts from your library when starting new tasks
Proactive	Offers to capture patterns when it notices you iterating

Proactive Detection

Requires setup: Add the proactive snippet to your CLAUDE.md — see Setup.

Promptly notices when you're in an exploratory pattern:

• 2+ rounds refining the same output (format changes, restructuring, adding sections)
• Pivots that reveal real intent ("actually, what I really want is...")
• Satisfaction after iteration ("yes, that's it", "perfect", "exactly what I need")
• Reuse signals ("I'll use this template going forward")

When detected, it offers to capture — no need to remember to run /promptly.

User Context Aware

If you have a CLAUDE.md or user profile, Promptly adapts distilled prompts to match your:

• Communication style
• Preferred frameworks
• Domain vocabulary

Installation

curl -o ~/.claude/commands/promptly.md \
  https://raw.githubusercontent.com/hazzap123/promptly/main/promptly.md

Or clone and copy:

git clone https://github.com/hazzap123/promptly.git
cp promptly/promptly.md ~/.claude/commands/

Setup

1. Create a prompt library

Start from scratch:

# prompt-library.yaml

prompts: []

Or copy the example library to get started immediately:

curl -o prompt-library.yaml \
  https://raw.githubusercontent.com/hazzap123/promptly/main/examples/prompt-library.yaml

This includes four ready-to-use prompts: Picard mode (truth over validation), belief reframing, content review, and strategic landscape mapping.

Place it in your project root, or wherever you prefer. If you use a different location, note it in your CLAUDE.md — Promptly will discover it automatically.

2. Enable proactive mode (recommended)

The /promptly skill only loads when you invoke it. For Claude to proactively detect prompt-worthy patterns during normal conversation, add a lightweight snippet to your CLAUDE.md.

Global (works in every project — recommended):

# Append the snippet to your global CLAUDE.md
sed -n '/<!-- promptly:proactive-start -->/,/<!-- promptly:proactive-end -->/p' \
  proactive-snippet.md >> ~/.claude/CLAUDE.md

Per-project:

# Append to a specific project's CLAUDE.md
sed -n '/<!-- promptly:proactive-start -->/,/<!-- promptly:proactive-end -->/p' \
  proactive-snippet.md >> /path/to/project/CLAUDE.md

Or manually copy the snippet from proactive-snippet.md. See that file for the full snippet content.

This tells Claude when to suggest /promptly — the skill itself handles the rest.

Without this step, proactive mode won't trigger and you'll need to remember to run /promptly manually after exploratory conversations.

Usage

After exploring a problem:

/promptly

Analyzes your conversation and offers to save a distilled prompt.

Check for existing prompts:

/promptly match

Searches your library for prompts matching your current context.

Example: Distill

/promptly

Analyzing your conversation...

You explored code review, iterating through:
- Initial ask: "review this PR"
- Added: security focus
- Refined: severity levels for issues
- Added: summary with recommended action

Here's your distilled prompt:

---
Review [CODE/PR] with focus on:

1. **Critical** — Security vulnerabilities, data loss risks, breaking changes
2. **Important** — Bugs, performance issues, error handling gaps
3. **Suggestions** — Style, naming, simplification opportunities

For each issue:
- File and line reference
- What's wrong and why it matters
- Concrete fix suggestion

End with: **Recommendation** — Approve / Request changes / Needs discussion
---

Triggers: "review this", "code review", "check this PR"

Add to library as 'code-review'? [Y/n]

Example: Proactive Offer

User: "Actually, can you add a section for open commitments?"

Claude: [Provides updated format]

User: "Yes, that's exactly what I need."

---
This looks like a prompt pattern worth capturing.

What I noticed:
- Initial ask: "prep me for meeting with X"
- Added: relationship history
- Added: open commitments
- Final form: structured prep doc

Want me to distill this as a reusable prompt?

[Yes, capture it] [No, just continue]
---

Library Format

Prompts are stored in YAML:

prompts:
  - id: code-review
    name: "Code Review"
    description: "Structured PR review with severity levels and actionable feedback"
    prompt: |
      Review [CODE/PR] with focus on:

      1. **Critical** — Security vulnerabilities, data loss risks, breaking changes
      2. **Important** — Bugs, performance issues, error handling gaps
      3. **Suggestions** — Style, naming, simplification opportunities

      For each issue:
      - File and line reference
      - What's wrong and why it matters
      - Concrete fix suggestion

      End with: **Recommendation** — Approve / Request changes / Needs discussion
    triggers:
      - "review this"
      - "code review"
      - "check this PR"
    created: YYYY-MM-DD

Customization

Library Location

Default: prompt-library.yaml in the project root.

Override by adding a line to your CLAUDE.md:

Prompt library: ./path/to/prompt-library.yaml

User Context

Promptly reads from CLAUDE.md and user-profile.yaml (if present) to personalise prompts. Document your:

• Communication preferences
• Frameworks you use
• Domain vocabulary

Distilled prompts will incorporate these automatically.

Proactive Sensitivity

Edit the <proactive_detection> section to adjust when Promptly offers to capture. Add or remove trigger signals based on your workflow.

Security

• Local only — Reads and writes files only within your project directory
• No network access — Cannot transmit data externally
• No code execution — Cannot run shell commands (Bash not in allowed-tools)
• Plain text storage — Prompt library is stored as YAML. Avoid capturing conversations containing credentials, API keys, or secrets
• User context stays local — Reads CLAUDE.md for personalization but doesn't store or expose its contents

Allowed Tools

allowed-tools:
  - Read
  - Write
  - Edit
  - Glob
  - Grep
  - AskUserQuestion

No Bash, no web access, no subagent spawning.

Requirements

• Claude Code CLI

Contributing

Issues, prompt examples, and improvements welcome. See CONTRIBUTING.md.

License

MIT

---

Turn exploration into assets. Build your prompt library over time.