skillify

name	skillify
description	Capture the current session's repeatable process into a reusable SKILL.md file
when_to_use	Call at the end of a process you want to capture and reuse. The agent reviews the conversation to extract a reusable skill definition.
argument-hint	optional: description of the process you want to capture
context	inline
disable-model-invocation	true

Capture this session's repeatable process as a SKILL.md file.

$ARGUMENTS

Round 1: Understand the process

Review the current conversation and answer:

1. What is the goal of this process? (one sentence)
2. What is a good skill name? (lowercase, hyphenated, e.g. review-pr)
3. What triggers this process? When should someone invoke it?
4. What does success look like?

Present these to the user and ask for confirmation or corrections before continuing.

Round 2: Design the skill

Once the goal is confirmed, determine:

1. Execution mode: inline (expands into the main conversation) or fork (runs in an isolated sub-agent)?
   • Use fork when the process is self-contained and should not consume the main conversation's context.
   • Use inline when the process needs to interact with the ongoing conversation.
2. Arguments: does the skill take user-supplied arguments? If so, what are they?
3. Allowed tools: which tools does this process need? List only what is actually required.
4. Save location: project-level (.claude/skills/<name>/SKILL.md) or user-level (~/.claude/skills/<name>/SKILL.md)?

Round 3: Write the SKILL.md

Produce a complete SKILL.md with:

• Frontmatter: name, description, when_to_use, argument-hint, context, allowed-tools
• Body: clear, numbered steps derived from what actually happened in this session
• $ARGUMENTS placeholder where user input should be inserted
• Each step should be atomic and verifiable

Show the draft to the user before writing.

Round 4: Save and confirm

1. Create the directory: mkdir -p <save-location>/<name>/
2. Write the file: <save-location>/<name>/SKILL.md
3. Confirm: "Skill /<name> is now available. Invoke it with /<name> [args]."

SKILL.md template

---
name: <name>
description: <one-line description>
when_to_use: <when to invoke this skill>
argument-hint: "<what args to pass>"
context: <inline|fork>
allowed-tools:
  - bash
  - read_file
---

<clear description of what this skill does>

$ARGUMENTS

## Step 1: <first step>

<instructions>

## Step 2: <second step>

<instructions>