> ## Documentation Index
> Fetch the complete documentation index at: https://docs.docwriter.org/llms.txt
> Use this file to discover all available pages before exploring further.

# Week of May 18, 2026

> Agent image reads, system prompt in transcripts, reply-only comments, and a new-file fix.

## New features

* **Agent can read images.** The agent's `Read` tool now accepts image
  files (PNG, JPEG, and friends) and ingests them natively, so you can
  point it at a screenshot or figure on disk instead of pasting it into
  chat. See [Steering the agent](/agent/steering-the-agent).
* **System prompt in the session transcript.** The session viewer's
  **System** card now shows the full system prompt the agent ran with,
  so you can see exactly what context — including your rules — was in
  play. See [Observability](/agent/observability).

## Updates

* **Rules are inlined in the system prompt.** Your writing rules are now
  baked into the agent's system prompt under "Rules to obey" instead of
  being appended to each user turn. The agent has the full rule list on
  every request, and the rules show up in the transcript. See
  [Steering the agent](/agent/steering-the-agent).
* **Agents reply, they don't start threads.** The agent can only reply
  on comment threads you've opened, not start new ones on its own. Use
  comments to drive a back-and-forth instead of getting unsolicited
  threads. See [Comments](/agent/comments).

## Bug fixes

* Creating a new file no longer carries over state from the previously
  active tab. New files now open empty as expected.