# The prompt

Paste this, then paste ten messages. Either the last ten in your inbox, or ten from the sample mailbox
if that is the safer answer for you.

Fill in the context block first — that is `03_ten_line_context_block.md`, and it is the part that does
the work.

---

```
You are triaging my email. Below is a short block about my work, then a set of
messages. Follow the three rules exactly.

=== WHAT I AM WORKING TO ===
<paste your ten lines here>
=== END ===

RULE ONE — QUOTE, DO NOT PARAPHRASE
For every item you give me, include the exact sentence it came from, copied
word for word from the message. Do not summarise it, do not tidy it, do not
correct its grammar. If you cannot quote a sentence that says it, do not
report the item.
Ignore quoted reply history and signature blocks entirely. Only use text the
sender wrote in this message.

RULE TWO — "NOTHING IS BEING ASKED OF ME" IS AN ALLOWED ANSWER
For each message, it is correct and expected to answer "nothing is being asked
of me." Do not find something in order to be useful. A message that is purely
informational, or addressed to a group with no action for me, should be
reported as no action.

RULE THREE — CONTRADICTION CHECK
Separately from the above, ask of every message: does anything here contradict
what I already believe about my work, as described in the block at the top?
A date moving, a dependency slipping, a person leaving, a number that does not
match. Report these even when nothing is being asked of me. Quote the sentence.

OUTPUT
For each message, in this order:

  MESSAGE <n> — <sender>, <subject>
  ACTION: <what is being asked of me, or "nothing is being asked of me">
  QUOTE: "<the exact sentence>"
  CONTRADICTS: <what it contradicts in my block, or "nothing">
  QUOTE: "<the exact sentence>"

Then, at the end:

  ASKS OF ME: <count>
  CONTRADICTIONS: <count>
  NO ACTION: <count>

=== MESSAGES ===
<paste ten messages here>
```

---

## Why each rule is there

**Rule one changes what review is.** Without the quote, checking means re-reading the thread, which
nobody does. With it, checking means reading one sentence and asking whether it says what the model
claims. Two seconds, and people actually do it.

The addition about quoted history and signatures is the difference between a safeguard and a
decoration. Models paraphrase. A quote that reads perfectly can be a tidy rewrite of something slightly
different, or a line from the reply history underneath that was settled last week. **When you check,
search your mail for the quoted string. If it is not there word for word, the item is not real.**

**Rule two stops invented work.** A model that has not been given permission to say nothing will
usually find something, and that is worse than missing one. A missed message leaves a gap, and gaps
make themselves felt. An invented action item leaves a list that looks complete, and you will work
through it.

**Rule three is the one that catches the cabinets** — the sentence in paragraph three of a friendly
update that asks nothing of you and makes your plan wrong. No rule you could write catches it, because
there is nothing to match on. It only works if the block at the top is filled in.

## When you look at the output

Count the items it says need action. Read the quoted sentence under each one. Does that sentence say
that?

Then count the other direction, which nobody tells you to do: **how many did it give you that did not
matter?** A triage that surfaces forty things to catch the two real ones has not helped you.
