feat(ask me): the question renders as markdown, and agents are told to write it

[L422Y]

What & why

An ASK ME question was printed as raw text. Agents were creating very hard to read questions, without any formatting, so I prompted so that they created markdown formatted questions, which was better, but the ASK ME view didn't support Markdown, so a card that said **Apply** or wrapped a path in backticks put
those characters on screen, and a question with three numbered options arrived as a
paragraph of punctuation. The question is the one thing on that card a human has to
read carefully, and it was the worst-formatted text in the app.

It now goes through the same MarkdownPreview the file preview and the IDE already
use, via a new card variant: it inherits the host card's mono face and size instead
of imposing the document type scale, and drops the 72ch measure and page padding so a
one-line question stays one line. The task detail's Q&A trail renders the same way,
answers included, since that is where the "view earlier answers" link lands.

Rendering it is only half the fix. Nothing had ever told the agents to format these,
so the orchestrator prompt and PROTOCOL.md now do, and PROTOCOL.md gains the ASK ME
section it never had (the humanQA contract was only documented in a source comment).
Shipping the renderer without that leaves the feature inert, which is why both are here.

Two details worth flagging for review:

• No new dependency. react-markdown and remark-gfm were already in the tree.
  CommonMark folds a single newline into a space, which would have turned every existing
  plain-text question into a run-on paragraph, so the card variant runs a twelve-line
  local remark plugin that splits text nodes on \n into break nodes. That is what
  remark-breaks does, and it did not seem worth a package. Code blocks are code nodes
  with a raw value, so their newlines are untouched.
• Still no rehype-raw. Agent-written markdown is untrusted and the existing file
  header says to keep it that way. Raw HTML in a question renders as text.

Type of change

• Bug fix
• New feature
• Refactor / cleanup
• Docs
• Build / CI

Evidence

Both shots are the same view, same 860x800 viewport, same dark theme, same two task
cards, differing only by this commit. The first card is a question written in markdown;
the second is deliberately plain prose with single newlines, to show that a question
with no markdown in it renders exactly as it did before.

Before

After

How I tested it

• OS: macOS (Darwin 25.3), Node 24.

• Steps:

  • npm run typecheck (both projects), npm run test:focused (552 pass), npm run build.
  • For the screenshots: mounted the real AskMeTab in a throwaway page outside the repo,
    against a stubbed window.cth returning a fixed four-task ledger, with the app's own
    tokens.css and global.css and the same Google fonts. Shot it in Chromium at one
    fixed viewport on main, then on this branch, changing nothing but the checkout. It is
    the real component and the real stylesheet, not the whole Electron window, so the
    Command Center tab bar is not in frame.
  • Checked the soft-break plugin against real mdast from mdast-util-from-markdown: a
    soft newline inside a paragraph becomes a break, and the newlines inside a fenced
    code block survive.
  • Rendered PROTOCOL_MD out of its template literal to confirm the escaped backticks
    and the fenced JSON block come out right in the file agents actually read.

What I have not done is run it in the packaged app against a live hive.

Discord (optional)

Discord:

Checklist

• Before and after evidence is attached above, under both headings.
• npm run typecheck passes.
• npm run test:focused passes.
• npm run build succeeds.
• This PR is one change. Unrelated fixes belong in their own PR.
• I read the diff myself before opening this, and there is no debug output,
  commented-out code, or unrelated formatting churn in it.
• Any new UI derives from DESIGN.md / tokens.ts — no ad-hoc colors,
  spacing, or fonts. The card variant adds no colors; every size in it is
  relative, so the host card sets the scale.
• If I added art, it's my own or compatibly licensed, and listed in
  ATTRIBUTION.md. (No art.)

[github-actions]

✅ Evidence received. Before and after are both attached. Thanks — this is what makes a PR reviewable in one pass.

[L422Y]

Unrelated to the diff, but worth knowing since it cost me a round trip: the PR evidence check fails when the image sits a blank line below its heading, which is exactly what GitHub's drag-and-drop produces.

The section regex in pr-evidence.yml uses the m flag, so $ in (?=\n#{1,6}\s|$) matches at the first end-of-line rather than end-of-string. The lazy group then stops immediately and the section captures an empty string:

section("### Before\n\n![a](x.png)")  // "" -> fails
section("### Before\n![a](x.png)")    // "![a](x.png)" -> passes

The failure mode is confusing rather than obvious: you get the near-miss message telling you both assets are in one place, when they are already under their own headings. Dropping the m flag, or using (?=\n#{1,6}\s|$(?![\r\n])), fixes it. Happy to open that as its own PR if it is useful.

EDIT: looks like #318 will address it

[chaitanyagiri]

Shipped in v0.4.6, and it is called out in the release notes.

Your commit 9edf34ea is in. The ASK ME card renders markdown, the Q and A trail in the task detail renders too, and the orchestrator prompt now asks agents to format what they write. Raw HTML still renders as text, which is the part that keeps agent written markdown safe to show.

This closes rather than shows as merged because the commit went in through the 0.4.6 release branch instead of this PR's branch.

https://github.com/chaitanyagiri/munder-difflin/releases/tag/v0.4.6

Thank you.