---
name: forget
description: Drop what was raised in this conversation, for good. Use only when the user invokes it and names what to forget.
license: CC0-1.0
argument-hint: "[what to forget]"
disable-model-invocation: true
---

Forget the target the user named when invoking this skill.

## What can be forgotten

Anything you produced that the user did not ask for: pending tasks, concerns and
risks, assumptions, inferred intent, options the user did not take, observations,
conventions you decided on your own, abandoned approaches.

## Classify first

Decide which one the target is. Do not ask the user.

- **Done** — it was handled. A task, a follow-up, a next step.
- **Accepted** — it stands, and the user is not acting on it. A concern, a risk,
  an option not taken.
- **Invalid** — it was wrong. An assumption, an inferred intent, a misreading, a
  convention you invented, a discarded attempt.

## Then

For all three: never raise it again. Not as a question, not as a reminder, not as
a caveat, not in passing. The user will bring it up if they want it.

For **invalid**, additionally: it is out. Do not reason from it, do not build on
it, do not let it narrow what you consider. Anything you concluded from it is
void — reach those conclusions again without it, or drop them.

## Rules

- **Touch nothing else.** The user drew the boundary. Do not widen it.
- **Do not stop to confirm.** If the target is identified, carry it out.
- **Do not argue.** Not before, not after, not one clause of hedging.

Ask only when the target cannot be identified, and ask only what it refers to.

## Reporting

One line per target: what it was, and which of the three it was treated as.

Nothing else. Not why, not what you considered, not what you kept.