Skip to content

Professional register, compact: hints on screen, not documentation - #84

Merged
vijay-tylertech merged 1 commit into
mainfrom
fix/professional-compact-copy
Aug 29, 2026
Merged

vijay-tylertech merged 1 commit into
mainfrom
fix/professional-compact-copy

Conversation

@vijay-tylertech

Copy link
Copy Markdown
Contributor

"'the rest is buttons' is casual talk""keep things as compact as possible with a professional tone""Do NOT put 'documentation' on screens. Assume just a hint is necessary."

The named example

Publishing needs an assistant for the knowledge files; the rest is buttons.
→ Requires an AI-enabled terminal with access to the affected files.

Used your wording with the grammar corrected (Publishing requireRequires) and tightened to a hint.

Measured on default-visible copy only

Anything inside a disclosure is opt-in and doesn't count as "on screen":

before after
eval opt-in 292 164
reset unsaved edits 275 138
transcript intro 287 150
summary intro 152 89
saves empty state 95 25

Register

Worth knowing                  → Reference
Something went wrong?          → Recovery
Show me exactly what changed   → View pending changes
Park a half-finished review    → For incomplete reviews
Saves what was typed. Decides nothing.
                               → Saves entered text. No verdict recorded.
Hands it over — the decision is given up.
                               → Transfers the decision to another owner.
best done off-hours            → Best outside business hours
a hint not a verdict           → indicative only
A review with nothing to fix is still worth sending.
                               → Reviews with no changes should still be submitted.

Two notes on changing things that weren't mine

The transcript intro banner was your wording, dictated verbatim earlier. Compacted here on the strength of "across the board" — and it had to change anyway for a factual reason: it told the reader to click Mark reviewed & next, a label that no longer exists. That button now reads No changes & next or Changes suggested & next depending on the form. Say the word if you want the longer version back.

Field help behind the ⓘ icons is untouched. It's opt-in, it's where long-form explanation belongs, and moving it on-screen is what the instruction forbids.

https://claude.ai/code/session_01VVNYqGza2dFdeWFX7qXnwb

"add a professional tone to messages - 'the rest is buttons' is casual talk";
"Do this across the board, keep things as compact as possible with a professional
tone"; "Do NOT put 'documentation' on screens. Assume just a hint is necessary."

The named example, using the reviewer's own wording with the grammar corrected:

  Publishing needs an assistant for the knowledge files; the rest is buttons.
  -> Requires an AI-enabled terminal with access to the affected files.

Applied to every string still carrying a casual register or explaining itself at
length. Measured on DEFAULT-VISIBLE copy only - anything inside a disclosure is
opt-in and does not count as "on screen":

  eval opt-in            292 -> 164 chars
  reset unsaved edits    275 -> 138
  transcript intro       287 -> 150
  summary intro          152 ->  89
  saves empty state       95 ->  25

  Worth knowing              -> Reference
  Something went wrong?      -> Recovery
  Show me exactly what changed -> View pending changes
  Park a half-finished review  -> For incomplete reviews
  Saves what was typed. Decides nothing. -> Saves entered text. No verdict recorded.
  Hands it over - the decision is given up. -> Transfers the decision to another owner.
  best done off-hours        -> Best outside business hours
  a hint not a verdict       -> indicative only
  A review with nothing to fix is still worth sending.
                             -> Reviews with no changes should still be submitted.

TWO NOTES ON WHAT I CHANGED THAT WAS NOT MINE.

The transcript intro banner was dictated verbatim by the reviewer earlier. It is
compacted here on the strength of the later "across the board" instruction, and it
also had to change for a factual reason: it told the reader to click "Mark
reviewed & next", a label that no longer exists - that button now reads "No
changes & next" or "Changes suggested & next" depending on the form.

Field help behind the ⓘ icons is untouched. It is opt-in, it is where the
long-form explanation belongs, and moving it on-screen is what the instruction
forbids.

Claude-Session: https://claude.ai/code/session_01VVNYqGza2dFdeWFX7qXnwb
@vijay-tylertech
vijay-tylertech requested a review from a team as a code owner August 29, 2026 15:55
@vijay-tylertech
vijay-tylertech merged commit b3534ee into main Aug 29, 2026
1 check passed
@vijay-tylertech
vijay-tylertech deleted the fix/professional-compact-copy branch August 29, 2026 15:56
vijay-tylertech added a commit that referenced this pull request Aug 29, 2026
Adds a collapsed History section under the open change requests on /prs, listing
merged requests from THIS flow in both the knowledge repo and Blueprint, each row
linking into the existing diff viewer rather than out to GitHub.

Provenance is the hard part, and the two repos have opposite problems: the
knowledge repo carries its own machinery work alongside review batches (81 of its
84 merged requests are machinery), and Blueprint is a shared docs repo with a lot
of unrelated traffic. So a request qualifies on EITHER of two markers:

  * the `onetyler-review` LABEL, applied at creation on both repos. Authoritative,
    and visible on GitHub itself rather than only through this tool.
  * the BRANCH PREFIX -- `review/` and `kb-review/`. This covers the back-catalogue,
    since the label only exists from today. `headRefName` survives the branch being
    deleted at merge (verified on #84, whose branch 404s on the branches API while
    the PR record still reports its name), so it stays readable indefinitely.

Either alone was insufficient: label-only would have shipped an empty history,
prefix-only leaves nothing identifiable on GitHub.

Tagging is never fatal. Labelling Blueprint may exceed a contributor's rights, and
the prefix keeps the request in the history regardless -- so a failure is reported
and the request is not re-opened. The label is created lazily, when a request is
opened, so viewing a dashboard never writes to Blueprint.

Also, in the diff viewer:

  * `?full=1` lifts BOTH truncation caps, reached from the truncation notice itself.
    The caps stay on by default, but "open it on GitHub" was the only escape from
    them, which made them a dead end on exactly the large requests worth reading.
  * Refresh now carries `repo` and `full`. It did neither, so refreshing a Blueprint
    diff silently fetched the KNOWLEDGE request with the same number.
  * A merged request said "closed", which reads as abandoned. It now says merged,
    with the date and who merged it.

The three historical review batches (#18, #31, #47) were labelled by hand so the
label is authoritative from the start rather than only for future requests.

Verified: the filter picks exactly those 3 of 84 merged requests and no duplicates
when both markers match; both caps confirmed lifted by `full=1` against forced
limits (12 lines/3 files on #18: 12 -> 323 lines, 3 -> 28 files); a bad Blueprint
number renders an error page rather than crashing; contrast green.

Claude-Session: https://claude.ai/code/session_01VVNYqGza2dFdeWFX7qXnwb
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant