How to iterate effectively in CharmIQ — the right places to prompt, how to recover from bad output, and when to start fresh instead of pushing forward.

Where Output Goes Depends on Where You Ask

There's one mechanic that trips up almost every new CharmIQ user: 

where you prompt determines what the Charm can do.

When you tag a Charm inside the main document, it appends. It writes below the cursor. It does not go back and revise what's already there. That's by design — the document grows forward as you work, and each Charm response becomes part of the permanent record.

. Tag your Charm there, ask it to update or rewrite specific content above, and it will reach back into the document and modify it in place.

main doc = add new things. Comments = fix existing things.

Once a document runs past a few pages, it gets genuinely hard to tell your words from the Charm's output. Lead every prompt you write with 

 (backslash + u + space — your name will appear). It marks your text visually so you can scan the document and immediately find your questions, decisions, and directions.

There's a second reason to do this: the Charm sees your markup too. When it reads the document, the 

 markers help it understand the conversation flow — what's yours, what's its output, where the work is heading.

Make it a habit from day one. Every time you write something in the doc, lead with 

When a Charm gives you something wrong or incomplete, the instinct is to prompt it at the bottom: 

Prompting below bad output traps you in a correction loop. The Charm reads everything above it — including its own wrong answer — and tends to justify or tweak rather than genuinely restart. You end up with a longer document and the same problem.

Scroll back up above the bad output. Add context, clarifications, or corrections 

 the prompt that generated the bad result. Then re-run. The Charm now sees corrected context first and generates something genuinely new, not a defensive variation on the original.

If the bad output is in the way, add a note like:

\u Note: ignore what follows — here's what I actually need...

Then re-prompt fresh. This one habit separates people who iterate efficiently from people who argue with their Charms.

When you're reading through Charm output for the first time, annotate immediately. Don't plan to do it on a second pass. Annotation fidelity drops sharply with cognitive fatigue, and a second pass rarely happens at the same quality.

This isn't just for your own benefit. The Charm sees your annotations as part of the document on every subsequent call. When you follow up, it already understands your preferences — what landed, what didn't — without you having to re-explain.

There's a point in every working document where continuing in the same space starts working against you. You've accumulated dead ends, exploratory tangents, corrections — the document has done its job, but it's also gotten noisy.

The temptation is to summarize within the document and keep going. Usually, that's the wrong move.

When a phase is genuinely complete, close it. Open a new workspace. Drag the finished document in as Reference Material. Start the next phase with only the 

 of the previous phase visible — not the mess that produced it.

The Charm gets a cleaner signal. You get a cleaner surface. And the record of how you got there is preserved in the original document if you ever need it.

 if the document has done its job, graduate from it. Don't summarize it.

Reference Materials aren't a set-it-and-forget-it feature. They're an active tool you should be adding and removing throughout a working session.

Too much reference material dilutes the Charm's focus. When it's trying to reconcile five documents at once, you get averaged output — responses that try to account for everything and land cleanly on nothing.

Think of Reference Materials as tools you pick up and set down, not shelves you keep permanently loaded.

 — How Charms work, how to invoke them, and how to configure them.

 — Designing Charm personas that produce sharper output.

 — Sequencing multiple Charms across a single project.

Where to prompt, how to recover from bad output, and when to stop pushing forward and start fresh.

Working Smarter

Where Output Goes Depends on Where You Ask

Mark Your Own Text

Recovering from Bad Output

Annotate as You Read

Knowing When to Start a New Document

Managing Reference Materials Dynamically

Patterns

Next Steps