← All free resources

Turn long explanations into the short version

Copy this skill into Claude, ChatGPT, or your AI agent to reduce a long explanation to the issue and the decision in plain prose.

How to Use It

Free to use, share, and modify.

SKILL.md · Condense~6 KB · Markdown
---
name: condense
description: When the user wants a wall of explanation reduced to the issue and the decision, in two short paragraphs of plain prose. Use when the user says "condense," the slash-command version of condense, "shorter," "tighten this," "give me the short version," "the version I can send," or when a long analysis has to be handed to someone who was not in the conversation. Invoked with no argument, it condenses the immediately preceding block of explanation.
---

# Condense

Reduce a long explanation to two short paragraphs: what the issue is, and what was decided. Write for a reader who was not in the conversation and will not scroll.

## What gets condensed

With an argument, condense the text, file, or thread named. With no argument, condense the immediately preceding wall of text in your own output. Do not ask which one is meant. The last long block is the target.

Output to the conversation. Do not write a file unless asked.

## The shape

Two paragraphs, one blank line between. No headers, no bullets, no bold, no closing question.

Paragraph one is the issue: what we wanted, what reality showed, and why the obvious fix does not work. Lead with the want, never with background.

Paragraph two is the decision: what we are doing instead, what we are deliberately not doing, and the cost of anything deferred.

If nothing has been decided yet, paragraph two is the recommendation plus the one thing it waits on. Never invent a decision to fill the slot.

## The budget

Aim at 400 characters. Shorter is better.

This is a pull, not a cap. 300 beats 400. Going past 400 is fine only when the extra characters carry something from the never-cut list below. Nothing else buys length, and "this is important context" in particular does not.

Never pad toward the number either. If the issue and the decision fit in 200 characters, answer in 200.

When you are over, cut a claim. Do not reflow sentences to save characters: reflowing holds the count roughly steady and costs you the plainness, which is the thing that made it readable.

## Keep

- The numbers. 18 flags, 6 flags, 78 percent, four minutes. Numbers are what let a reader who was not there check the claim.
- Exactly one concrete example: the decisive one. The case that made the problem undeniable, not the one that is easiest to explain.
- The structural reason, in a single clause, folded into another sentence if that is what fits. Why this is not a tuning problem.
- Anything deferred, broken, or knowingly imperfect, with its cost.

## Cut

- Options considered and rejected. The decision implies them.
- Process narration: how the analysis went, what you checked, what surprised you.
- Restatement. The same point made again in a second register is the largest single source of length.
- Openers and hedges. "This is worth stopping on." "It is worth noting." "So both rules are wrong in opposite directions."
- Secondary examples. Two illustrations of one phenomenon is one too many.
- The offer to do the next thing. If a next step matters it is a decision, so state it as one in paragraph two.

Restatement and secondary examples are the two doors length comes back through. Check them first when a draft runs long.

## Never cut

A deferral, a known cost, or something that is broken. A condensation that reads cleaner because it quietly dropped the bad news is wrong, not short. If one of these will not fit, go past 400 rather than lose it, and do not remark on having done so.

## Voice

Plain declarative prose. No em dashes: use a colon, a period, or restructure. No semicolon chains standing in for the sentences you cut. Name things the way the reader names them, not the way the code does.

## Check before answering

1. Two paragraphs. Not one, not three.
2. Every number traces to the source. Condensing is not the moment to round or estimate.
3. Nothing deferred or broken went missing.
4. Read paragraph one alone. If it does not state a problem, it is background, and background is not the issue.
5. Anything past 400 characters is carrying a never-cut item. If it is not, cut it.

## Example

Before, about 1,900 characters: two failed rule thresholds walked through in turn, three example clients, a structural argument, four bullets of what to keep from an earlier proposal, a constraint on one UI state, and a closing offer to write it up.

After, 524 characters:

> Wrong-account mappings cannot be detected from names: similarity cannot tell one business spelled two ways from two spelled alike. A loose rule flags 18, nearly all harmless formatting. Tightened to 6, it goes silent on HealthPro Heritage mapped to HealthPRO Pediatrics, the one that matters.
>
> So no verdicts. We hide a connector label only when it restates the client name, 78 percent of them, and show the rest plainly. The broken-mapping state is deferred: it needs live data that drags a four-minute stall into the list.

Past 400 because of the last sentence, which is a deferral and its cost. Everything else was cut: the second and third example clients, each threshold walked through in turn, the four bullets restating the earlier proposal, and the closing question. What survived: every number, the one decisive mismatch, the reason no threshold works.

What this skill does

Two paragraphs

Turns a wall of explanation into the issue and the decision, with no headers, bullets, or trailing ask.

Keeps the point

Preserves the numbers, the decisive example, and any bad news that would change the reader's decision.

Cuts the slop

Removes process narration, repeated points, soft openers, extra examples, and options already rejected.

Sendable prose

Produces a short version that works for someone who was not in the conversation and will not scroll.

Use cases

  1. Turn a long analysis into the version a client or teammate can read quickly.
  2. Condense a decision memo into what changed and what the team is doing next.
  3. Shorten a Slack, email, or ChatGPT response without losing the bad news.
  4. Create a tight handoff for someone joining the conversation late.

How to use it

Copy the skill once, then add it to whichever AI you already use.

Claude

  1. Copy the skill above
  2. Paste it into a Claude Project or Skill
  3. Add the long explanation or rely on the previous message
  4. Ask: “condense this”

ChatGPT

  1. Open a GPT or project chat
  2. Paste this skill as instructions
  3. Paste the text to shorten
  4. Ask for the version you can send

Google Gemini

  1. Create a Gem
  2. Paste this skill into the Gem instructions
  3. Give it a long analysis or thread
  4. Ask for two short paragraphs

AI agent

  1. Save as SKILL.md
  2. Place it in skills/condense/
  3. Use it when summaries need to become decisions
  4. Keep the output in the conversation unless asked otherwise

Want sharper AI workflows?

We build practical skills, agents, and workflows for teams that need AI to make decisions clearer, not just generate more text.