prreviewbuddy 0.52.2 → 0.55.1

package/CHANGELOG.md

@@ -1,5 +1,104 @@
 # Changelog
 
+## 0.55.1
+
+**Load more on the Notifications page.** The list shows the newest fifteen, and Load more reveals
+fifteen more at a time, the same way each section of the Reviews page does. Mark all read still
+covers every notification, including the ones not yet shown.
+
+## 0.55.0
+
+Published with 0.54.0, which did not go to npm on its own: read both entries below.
+
+**Why this change?** The Overview now says why the change exists, between Author's context and
+What's changing, and how it knows:
+
+- **Stated by the author**, with a short quote from the description. The quote is checked
+  against the description the analysis read, and **From Author's context** takes you to it.
+- **Likely motivation**, when nothing states it but the change gives evidence: a test, an issue,
+  a commit message. The evidence is listed, with links to its files.
+- **Motivation unclear**, when knowing why would change how the change should be read. The
+  analysis asks the author one question, filed first on the Questions page under **Intent**, with
+  the same Mark answered control as every other question.
+
+Nothing is shown when the reason would not change the review, or for analyses made before this
+version. A claim that fails its check is never shown in the model's words: an unverified quote or
+an inference without evidence becomes "unclear" with a neutral sentence when the analysis asked a
+question, and is left out when it did not. The workspace says so under **What the harness
+corrected**.
+
+- The workspace's prompt now reads the pull request's description. A local branch's commit
+  subjects are labelled as commits rather than as a description.
+- Updating an analysis with new commits keeps its motivation, since commits show what changed
+  rather than why. The extension 0.25.0 shows the same Overview.
+- Once the Intent question is marked answered, the Overview says **Motivation was unclear when
+  analysed** and offers **Run review again**, since marking a question answered records no answer
+  for an update to use. A new analysis reads the description and conversation as they now stand.
+  The workspace page can now start that analysis itself, as `review --reanalyse` does.
+- Evidence follows one rule in both places. A changed file opens in the diff, and a real unchanged
+  file opens in the source view. A path the reviewed commit does not hold is removed and listed
+  under **What the harness corrected**. A path that cannot be checked, which is any path outside
+  the pull request in the extension, is shown as an unverified path rather than a link.
+- A quote counts as the author's only if it is in the text they wrote. A partly filled template's
+  comments, unticked checklist items and bare headings no longer count.
+- Tabbing into a closed Author's context preview opens it, so a focused link or Copy button is
+  never hidden under the fade.
+
+## 0.54.0
+
+**The Overview starts with the change, then gives the verdict.** The risk card used to come first,
+and its takeaway read like the pull request's description. The Overview now opens with
+**Author's context**, the pull request's description exactly as the author wrote it. Under it is
+**What's changing**, the analysis's summary of what the code does. The risk card, the counts and
+the suggested review path follow, as before.
+
+- Author's context appears only when the pull request has a description. A description that is
+  only an unfilled template counts as none, and so does a review of a local branch or commit.
+- Each analysis keeps the description it was made against. If the author edits it later, older
+  analyses still show what they read, and the next analysis reads the new text.
+- Long descriptions open as a preview with **Show full description**. All of it, checklists and
+  testing notes included, is one click away, and **Show less** folds it again.
+- The workspace now fetches the description in the same `gh pr view` call it already made. The
+  extension 0.24.0 shows the same Overview.
+
+## 0.53.1
+
+**The running KISS review has its normal layout back.** While a KISS review ran, its progress
+message sat in a rounded grey pill with the text pressed against its edges. It borrowed the styling
+of the new recommendation status labels, which shared its name.
+
+## 0.53.0
+
+**KISS recommendations are decisions now.** Each recommendation carries its status and two choices
+worded for whose code it is. Reviewing your own code, you choose **Plan to simplify** or **Keep as
+implemented**. Reviewing someone else's, you choose **Raise with author** or **Leave it**. Decided
+recommendations move into a **Decided** list below the open ones, in the order the analysis ranked
+them. Press the chosen button again to take a decision back. Decisions are kept across reloads, and
+the KISS rail counts what is still undecided. If a decision can't be saved, the card goes back as it
+was and offers **Try again**.
+
+- **Copy recommendations** (reviewing someone else's code): copies the recommendations you chose to
+  raise as Markdown ready to paste, with each one's title, why it matters, the simpler approach, where
+  to start and the draft comment. It is disabled until you choose one, while a decision is still
+  being saved, and whenever the branch has moved since the run. If a rerun fails, the earlier review
+  stays on the page and is checked against the branch by its own commit, so it is marked stale when
+  the branch has moved past it.
+- Each card also has **Copy**, **Ask Buddy**, **Open file** and, where the run wrote one, **Copy
+  draft comment**.
+- **Whose code is this?** is now part of running KISS, not part of the result. A finished review
+  shows its perspective in the header, beside **Run again**, which opens the choice with the last
+  answer already selected. The stale-branch notice offers the same **Run again**, and **Not now**
+  now folds that notice to one line instead of removing it.
+- **Complexity worth keeping** is collapsed on the KISS summary. New runs give each item a one-line
+  reason beside its title.
+- Fixed: a review finishing in the background could undo something you changed on the page while it
+  ran, such as a dismissed notice or an answered question.
+
+## 0.52.3
+
+**Selected cards keep their rounded corners.** In select mode, a ticked card's tint now fills the
+card's own rounded shape instead of a square behind it.
+
 ## 0.52.2
 
 **Continue reviewing is gone from the Reviews page.** The banner opened whatever was first in your