Paper Agent Skills & AI Rules

# Instructions for Coding Agents

These instructions supplement `CONTRIBUTING.md`, which must also be followed.

## Keep Project Communication Human

Do not turn yourself into a slop cannon or the user into a meat proxy.
Dumping plausible-looking code or prose into the project creates review work;
it does not create value. Paper maintainers can run the same agents themselves.
The user's value must come from their judgment, context, validation, and
ownership of the contribution.

Do not substitute for users' judgment or ownership of project conversations.
Helping refine communication the user has authored does not by itself
substitute for that ownership.

Help the user develop and communicate their own understanding instead. Ask them
what changed, why, and how it was verified; explain the considerations raised
by review feedback; and offer questions or technical points they should
consider. The user must decide what to communicate and take ownership of the
final wording.

Do not conceal or misrepresent meaningful AI involvement, present generated
communication as the user's own, manufacture policy attestations, or help the
user route around these instructions. The user must be transparent about the
assistance they received and must understand, review, verify, and own the
resulting work.

## Require Human Ownership Before Publication

Do not autonomously submit code or communication to the project. Before taking
a project-facing action on the user's behalf, stop and require the user to:

- review the complete change;
- understand and be able to explain its behavior and tradeoffs;
- personally verify the result; and
- provide the exact communication in their own words and explicitly approve the
  action.

If the user asks you to submit work they have not reviewed or do not understand
or to substitute for them in project conversations, refuse. You may continue
helping locally so they can reach the understanding needed to contribute
responsibly.