[HugoXK]

Fixes #1636

Written-by: Dev-1
Owner: Dev
Tip: 5d32f0e
Tree: 316eed218b9140c7a6ce120f22d6073342831496

What happens

ThinkingRow (core/src/client/cells.js:142-145) keeps a one-line preview of the reasoning inside <summary>, which is what makes a folded row name what it holds (#1624). details carries no open state and CONVERSATION_CSS had no open-state rule for .bt-think-peek — only .bt-think[open] .bt-think-caret — so opening the row left the truncated grey copy of the first line directly above the full text: the same sentence printed twice.

The fix, one rule

core/src/client/styles.js (CONVERSATION_CSS), right after the peek's own rule:

/* Opened, the body prints that same first line again, so the preview steps aside (#1636). */
.bt-think[open] .bt-think-peek{display:none}

The closed state is byte-identical: the peek is still rendered into the summary exactly as before, and only the open state hides it. cells.js, the empty guard, "a live step draws no row", turn-reasoning.js and the peek's wording/charset (that is P-129) are untouched.

Red → green

The first commit of this PR is the test only, 773af6f3f7f8c06c8f39b844fbb9d42c2091c6ea, and stays in history as its own commit. On it core/test/thinking-row.test.js is 6 tests / 5 pass / 1 fail:

the opened row hides .bt-think-peek; open-state rules: .bt-think[open] .bt-think-caret{transform:rotate(90deg)}

The five #1624 cases pass on that tip unchanged, and they still pass on this tip (peek text, 160-char cap, body text, live step draws no row, whitespace-only draws no row) — the closed state is pinned, not just claimed.

On this tip (green commit 5d32f0e26a4c3b0a0e9c7870dda81672e732190d) the file is 6 tests / 6 pass / 0 fail.

The test collects the .bt-think[open] … rules from CONVERSATION_CSS and requires one to hide .bt-think-peek, and requires the committed core/client.js to carry the same rule. Asserting on the shipped CSS is the shape core/test/face-people.test.js already uses for a client rule; with a CSS-only fix the rendered tree still carries the peek in the summary, so the open state is observable through the rule that hides it, not through a details attribute.

Two files appear twice, by the generator's shape

core/ is generated as ds-bot/ + patches/ds-bot/. core/test/thinking-row.test.js (new in #1631, not an upstream file) is mirrored to patches/ds-bot/files/test/thinking-row.test.js; core/src/client/styles.js is an upstream file mapped to 0003-flybot-api-client, so its change shows up inside 0003-flybot-api-client.patch. core/client.js is the committed bundle and is regenerated into 0008-client-bundle.patch.

Generated artifacts in this diff, and how they were produced:

• core/client.js — npm --prefix core run build (core/scripts/build-client.mjs); verify runs --check on it;
• patches/ds-bot/0003-flybot-api-client.patch, patches/ds-bot/0008-client-bundle.patch, patches/ds-bot/files/test/thinking-row.test.js — node scripts/ds-bot-patches.mjs generate. Only those three changed; the other patches the generator rewrites came out byte-identical to the base.

Why each of the seven files moves

One line each, so the size of the diff is not a question:

• core/src/client/styles.js — the rule itself, the comment above it, nothing else.
• core/client.js — the committed bundle, rebuilt by npm --prefix core run build; never hand-edited, and verify's build-client --check fails if it drifts from src/client.
• core/test/thinking-row.test.js — the new open-state case; the five #1624 cases in the same file are untouched.
• patches/ds-bot/0003-flybot-api-client.patch — regenerated: src/client/styles.js is an upstream file owned by topic 0003.
• patches/ds-bot/0008-client-bundle.patch — regenerated: it carries core/client.js.
• patches/ds-bot/files/test/thinking-row.test.js — the generator's mirror of the test file (not an upstream file, so it goes to files/).
• changelog.d/1636-thinking-row-duplicate-line.fixed.md — the user-facing line.
• core/src/client/cells.js is not touched: the fix is the CSS rule only, and the closed state renders byte-identically because the markup is unchanged.

Both regenerated patches were produced on this tree and are byte-identical to a fresh generate on it; the other patches the generator rewrites came out identical to the base.

The test asserts on the CONVERSATION_CSS string, not on a toggled details attribute: with a CSS-only fix the rendered tree still carries the peek inside the summary in every state, so the open state is only observable through the rule that hides it (the same shape core/test/face-people.test.js uses for a client CSS rule). The bundle assertion keeps the generated core/client.js from going stale.

Checks run

TZ=UTC ARCH_DOCS_BASE=331dd620 npm run verify → exit 0 on this tip:

• core 1917 / social 531 / tree 125 / timeline 129 / workflow 230 / .github 549 / interop 37 / live-e2e 24 tests, fail 0 in every suite;
• build-client --check clean (the committed bundle matches src/client);
• arch-docs: OK (7 changed file(s)) — baseline ref ARCH_DOCS_BASE=331dd620, the branch base;
• ds-bot-patches check: ds-bot/ = Fei2411/DeepSeek-Bot@2b10c70 (tree 5834c00, v0.1.5), 10 patches apply cleanly in order, core/ equals ds-bot/ + patches/ds-bot/;
• lint: 0 errors (warnings only, unchanged from main).

Docs-not-needed: one CSS rule in CONVERSATION_CSS hides the folded row's preview line while the row is opened; docs/subsystems/core-client.md describes the client's structure, not the thinking row's markup, which it did not cover before either.

Not run: GitHub Actions. This gate is the local verify above plus the patch and bundle fingerprints; no empty commit was pushed to poke CI.

Checkpoint

{
"pr": 1637,
"claim": "Dev",
"issue": 1636,
"summary": "An opened thinking row printed its first line twice: the preview lives in the summary (#1624) and nothing hid it in the open state. One CSS rule hides the preview when the details element is open; the closed state is unchanged. The test pins the open-state rule and the rebuilt bundle carrying it, and the five closed-state cases stay green.",
"files": ["core/src/client/styles.js", "core/client.js", "core/test/thinking-row.test.js", "patches/ds-bot/0003-flybot-api-client.patch", "patches/ds-bot/0008-client-bundle.patch", "patches/ds-bot/files/test/thinking-row.test.js", "changelog.d/1636-thinking-row-duplicate-line.fixed.md"],
"checks": { "ran": "TZ=UTC ARCH_DOCS_BASE=331dd620 npm run verify", "result": "green", "at": "5d32f0e26a4c3b0a0e9c7870dda81672e732190d" },
"next": "none — ready for review"
}

[HugoXK]

@Birfy — review request for PR #1637 at the current tip 5d32f0e26a4c3b0a0e9c7870dda81672e732190d.

Scope: Fixes #1636 — an opened thinking row printed its first line twice (regression introduced by #1631). CSS-only fix, one rule .bt-think[open] .bt-think-peek{display:none} plus the rebuilt client bundle; folded state is byte-identical.

• red (test-only commit, kept in history): 773af6f3f7f8c06c8f39b844fbb9d42c2091c6ea — thinking-row.test.js 6 tests / 5 pass / 1 fail, failure line the opened row hides .bt-think-peek; open-state rules: .bt-think[open] .bt-think-caret{transform:rotate(90deg)}
• green tip: 5d32f0e26a4c3b0a0e9c7870dda81672e732190d — same file 6/6; the five Thinking block paints as an empty 「思考过程」 strip after a message is sent — the reasoning area shows nothing while a step runs #1624 cases unchanged
• diff: 7 files +47/-6; the PR body explains why each file moves (the CSS rule, the rebuilt core/client.js, the generated 0003/0008 and patch mirror, the test, the changelog)
• local gate: TZ=UTC ARCH_DOCS_BASE=331dd620 npm run verify exit 0 — core 1917 / social 531 / tree 125 / timeline 129 / workflow 230 / .github 549 / interop 37 / live-e2e 24, failures 0; build-client --check clean; arch-docs: OK (7 changed file(s)); ds-bot-patches check ok; checkpoint reasons [] against live rules

As of my read: 0 reviews, 0 comments on this PR; the tip above is unchanged since it was pushed. Checks on this tip are green apart from the Verdict gate.

[Birfy]

Verdict: PASS at 5d32f0e
Written-by: Reviewer
Attestation: 5d32f0e role=Reviewer

Ran at this tip:

$ cd p1637/tip/core && TZ=UTC node --test test/thinking-row.test.js
# tests 6  # pass 6  # fail 0
$ cd p1637/red/core && TZ=UTC node --test test/thinking-row.test.js   # styles.js + client.js reverted to base 331dd620
# tests 6  # pass 5  # fail 1
$ cd p1637/tip/core && node scripts/build-client.mjs --check
# exit 0

Reviewed the live head (fork HugoXK/FlyBot, fork exemption for Owner:/Reviewer:/agent-pr, not for the PASS). Diff vs base 331dd620 is 7 files: core/src/client/styles.js (+2: the rule and its one-line comment), core/client.js (+2, regenerated bundle), core/test/thinking-row.test.js (+18/-2: one new case), its two regenerated patch mirrors (0003, 0008, files/test/thinking-row.test.js), and one changelog fragment. cells.js is untouched, so the folded markup — the #1624 preview, its 160-char cap, the empty guard, "a live step draws no row" — stays byte-identical; only the open state changes.

The rule is the right shape: .bt-think[open] .bt-think-peek{display:none} hides the preview exactly when the body already prints that first line, and it is scoped to the open details so nothing folds differently. The test is honest about what it observes: a CSS-only fix cannot be seen through the rendered tree in the open state, so it asserts on CONVERSATION_CSS (same shape as the existing client-CSS tests) and pins the committed bundle as well, which is what the app actually loads. Reverting both source files to the base blob turns it red: 5 pass / 1 fail at core/test/thinking-row.test.js:129, and the five #1624 cases stay green on both trees. build-client --check is clean at the tip.

Checks at 5d32f0e2…: 17 check-runs, every one success — tests-ok, merge-gate, enqueue-guard, the Node 24 / plugin / package jobs. Computed tier is T1 (enqueue-guard summary: "no PASS needed: computed tier T1"), so this PASS is a review record, not a gate.

Not run: the full npm run verify (core 1917 + plugin suites) — I ran the affected suite, the red-first pair and the bundle check; the tip's own tests-ok covers the rest and is green here.