process-improvement

The user wants to improve a process based on a bad or suboptimal AI interaction.

They may provide $ARGUMENTS as inline text, a file path, or a general description
of what happened. If $ARGUMENTS is empty or vague, ask them to paste or describe
the interaction that didn't go well.

## Step 1: Understand the material

Figure out what the user is reacting to. This could be:
- A snippet of AI output they didn't like
- A skill invocation that produced a bad result
- A general pattern they've noticed across multiple interactions
- A workflow step that feels clunky or wrong

If they gave a file path, read it. If they pasted text, work with that.
If they described something from memory, ask enough to understand the specifics.

Don't rush past this. You need concrete examples, not just vibes.

## Step 2: Diagnose the gap

Walk through the problem with the user. Help them articulate:

1. **What happened** - what the AI actually produced or did
2. **What they wanted** - what the ideal outcome would have looked like
3. **Where the gap is** - which of these root causes apply:
   - Bad or missing instructions (the prompt/skill didn't say what to do)
   - Wrong framing (the prompt asked for the wrong thing entirely)
   - Missing context (the AI didn't have what it needed)
   - Wrong tool for the job (should have been a different skill, manual step, etc.)
   - Scope mismatch (too broad, too narrow, too many steps at once)
   - Style/tone issue (output was technically correct but felt wrong)

Present your read of the situation and let the user react. Don't assume you
nailed it on the first try - this is a conversation.

## Step 3: Design the fix

Based on the diagnosis, propose a concrete improvement. This could be:

- **A prompt rewrite** - revised instructions or framing for next time
- **A skill edit** - specific changes to an existing SKILL.md file
- **A new skill** - if the gap suggests a missing capability
- **A workflow change** - doing things in a different order or splitti