prompt-engineering

Prompt Engineering

CRITICAL: Check existing project conventions before inventing new ones. Read CLAUDE.md, .claude/rules/, and existing skills to align with what's already established.

Skill Creation Workflow

Step 1: Analyze the Session/Request

Before asking questions, analyze what you already know:

• What repeatable process was performed or requested
• What the inputs/parameters are
• The distinct steps in order
• Success criteria for each step
• Where the user steered or corrected during the session
• What tools and permissions are needed

Step 2: Interview (4 Rounds)

Use structured questions. Don't over-ask for simple processes.

Round 1 — High-Level Confirmation

• Suggest a name and description. Ask to confirm or rename.
• Suggest goal(s) and specific success criteria.

Round 2 — Details

• Present the steps as a numbered list.
• Suggest arguments based on observed inputs.
• Ask: inline (current conversation) or forked (sub-agent with own context)?
  • Fork for self-contained tasks without mid-process user input
  • Inline when the user wants to steer mid-process
• Ask where to save:
  • This repo (.claude/skills/<name>/SKILL.md) — project-specific workflows
  • Personal (~/.claude/skills/<name>/SKILL.md) — follows you across repos

Round 3 — Per-Step Breakdown For each major step, if not obvious:

• What does this step produce that later steps need?
• What proves this step succeeded?
• Should the user confirm before proceeding? (especially irreversible actions)
• Are any steps independent and could run in parallel?
• What are the hard constraints?

Round 4 — Final

• Confirm when the skill should be invoked. Suggest trigger phrases.
• Ask for gotchas or edge cases.

Step 3: Write the SKILL.md

Use this template:

---
name: {{skill-name}}
description: {{one-line description}}
allowed-tools:
  {{list of tool permission patterns}}
when_to_use: {{detailed invocation triggers with example phrases}}
argument-hint: "{{hint showing argument placeholders}}"
arguments:
  {{list of argument names}}
context: {{inline or fork — omit for inline}}
---

# {{Skill Title}}
Description of skill

## Inputs
- `$arg_name`: Description of this input

## Goal
Clearly stated goal with defined artifacts or completion criteria.

## Steps

### 1. Step Name
What to do. Be specific and actionable. Include commands when appropriate.

**Success criteria**: What proves this step is done.