feat(mulmoterminal): Closing Summary and Header Chips clips, en and ja - #125
Merged
Merged
Conversation
… / #47) Two backlog clips from the 2026-09-10 shoot, English and Japanese. Closing Summary (#27) needs no setup beat: appendSystemPrompt is true by default, so what the clip shows is what a stock install does. The demo repos had it forced off, which is why it had never been filmed; that override is gone now and the rig matches the product default. Header Chips (#47) does need one, and gets it as a static config card -- chips are one of the four keys with no Settings page, so a viewer who only sees the payoff would look for a button that is not there. The chip shows ${task}, the worktree the cell is in. The built-in git chip already carries ahead/behind, so a chip repeating those reads as decoration; the worktree name is on no built-in, and the branch chip truncates it. Beat 1 of Header Chips is cut from 7.0s of the take, not 5.0s: the worktree cell only appears at 5.5s, and the first frame is what X shows. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MpSytuz8z3ZXuQ6vGNxTab
… it being written Three notes from the author, all about the clip not saying which thing it means. The setup beat was a code card. Asked how you actually write the chip, the card does not answer -- it shows the JSON without showing where it goes or what puts it there. It is now a frame from a real run: a cell asked for the chip, the app's own mulmoterminal-header skill loaded, posted the array through the live API and said what it wrote, keys and all. The same objection retired the run-menu code card in September. Nothing pointed at the chip, so "this one names the worktree" had no referent. It has an amber ring now -- a ring rather than a pulse because the chip is not something you press. And the click on the issue row had no pointer at all. It follows the recipe: the arrow travels in over 0.65s, the box flashes for a quarter second on arrival, and the press lands after both are gone. Two things the shoot settled and the clip does not show. The chip appears once the header re-resolves -- expanding the cell was enough -- so the restart that was meant to trigger it turned out to be unnecessary. It was also harmful: on a cell started from an issue, restart is refused with "this worktree already has a session" and the cell goes to disconnected. Worth a look upstream. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MpSytuz8z3ZXuQ6vGNxTab
…t an agent's report The previous setup beat was a frame of the agent's reply after it had written the chip. Asked why that screen was there, the answer was that it did not say: it showed the result -- Skill(...), POST /api/config -- with nothing a viewer would do themselves, and no sign of where it was asked from. The product already has the entry point. Settings, Header buttons and chips, "Set up header buttons...", with the line under it that answers the question outright: an agent does this for you, it asks a few questions and edits the config file itself. That screen is the beat now, with the pointer travelling to the button and a quarter-second box on arrival. It is not pressed; what happens after is the agent's conversation, which is a different clip. Built from a still. Settings is a fixed inset-0 modal, which reliably freezes a screencast, so the rig takes a PNG and the button's rect and the pointer is composited afterwards. Captured on 4.20.0. The clip went 7s at first and check-beat-fit called it tight against 6.5s of narration; the loop is 9s now rather than the line shorter. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MpSytuz8z3ZXuQ6vGNxTab
… press the button through to the request Two notes from the author on the previous cut. The issue-start beat cut from the issue list to a four-cell grid and stopped there, so nothing said which of the four was the worktree just created. The grid now holds for about a second, the camera pushes into the bottom-left cell, and the narration names the place: "Its worktree opens at the bottom left". The beat is pinned to 10s in the deck -- its length comes from that field, not from the audio, and at the old 7.5s the pushed-in frame was on screen for a fifth of a second. The Settings beat stopped at the pointer over "Set up header buttons...". The author wanted the rest: pressing it opens a confirmation, "Let an agent set this up?", and Start opens a terminal in the grid where Claude starts with the request already in. That is two beats now -- the button and the dialog, then the new terminal with /mulmoterminal-header landing, pushed in so the request reads. The dialog focuses Cancel by default. The pointer lands on Start with an amber box, because a frame of that dialog with nothing marked reads as Cancel chosen. The clip is 29.9s, over the 15-25s the clip format asks for. Nothing in it is padding; the length is what the two additions cost. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MpSytuz8z3ZXuQ6vGNxTab
ystknsh
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
MulmoTerminal の機能を紹介する短尺クリップ 2 本を、英語版と日本語版で追加します。どちらも新しい版で入った機能ではなく、以前からある機能（stock）です。発表者の 09-06 のデッキにある機能スライドから拾いました。
mt-closing-summary（0:15） — 各セッションの返答の末尾に「頼まれたこと・やったこと・やっていないこと」の 3 行が付く。別のセルへ行って戻ってきたとき、スクロールし直さずに読める。既定で有効（appendSystemPromptの既定は true）なので、設定ビートはありませんmt-header-chips（0:30） — ヘッダーに自作のチップを足せる。例として、そのセルが居る worktree の名前（${task}）を出すチップを使っています。issue 一覧の ▶ から始めたセルは 4 分割の左下に開き、最初からチップに名前が出る。チップの追加は Settings のボタンから行い、新しいターミナルでエージェントが設定する完成した動画
Header Chips 英語版:
mt-header-chips_en.mp4
Header Chips 日本語版:
mt-header-chips_ja.mp4
Closing Summary 英語版:
mt-closing-summary_en.mp4
Closing Summary 日本語版:
mt-closing-summary_ja.mp4
Items to Confirm / Review
flow.mp4）は、デッキのdurationを 10.0 に固定しています。7.5 のときは、寄った後の画が 0.2 秒しか映りませんでした。音声（英 6.3 秒・日 5.9 秒）より長いため、このビートの尺はdurationが決めていますcheck-beat-fit.pyは 4 デッキとも通過しています（最小の余裕は 1.10 秒。Header Chips のビート 2）User Prompt
制作メモ
mulmoterminal-videoskill 同梱のrecord-closing-summary.mjs/record-issue-worktree.mjs/record-settings-setup.mjs（この PR とは別リポジトリ）footage/2026-09-10/（Closing Summary はsummary-d、チップと issue 起動はwt-b）とfootage/2026-09-11/（Settings と新しいターミナル）appendSystemPrompt: falseで既定を上書きしていたので、09-10 に外して素の install と同じ状態にしています🤖 Generated with Claude Code
https://claude.ai/code/session_01MpSytuz8z3ZXuQ6vGNxTab