commandmate 0.44.2 → 0.44.3

package/dist/server/src/lib/polling/response-checker-extraction-steps.js

@@ -0,0 +1,670 @@
+"use strict";
+/**
+ * The steps split out of `extractResponse`, including the per-tool
+ * startup-screen defenses (Issue #3374 split from response-checker.ts).
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.findChromeStart = findChromeStart;
+exports.findRecentUserPromptIndexInFrame = findRecentUserPromptIndexInFrame;
+exports.isTurnComplete = isTurnComplete;
+exports.extractCompletedResponse = extractCompletedResponse;
+exports.extractPartialResponse = extractPartialResponse;
+const cli_patterns_1 = require("../../lib/detection/cli-patterns");
+const logger_1 = require("../../lib/logger");
+const thinking_constants_1 = require("../../config/thinking-constants");
+const logger = (0, logger_1.createLogger)('response-poller');
+const response_extractor_1 = require("../response-extractor");
+const startup_screen_1 = require("./startup-screen");
+const response_poller_constants_1 = require("./response-poller-constants");
+const response_checker_extraction_result_1 = require("./response-checker-extraction-result");
+/**
+ * How many rows from the bottom of a copilot capture may hold its status bar.
+ *
+ * `readCopilotStatusBar` stops at the first non-blank row from the end, and the
+ * capture handed to it here has already had its trailing blanks trimmed, so one
+ * row is enough in practice. The slack exists so the reader keeps working if that
+ * trim ever changes, without mapping a thousand rows through `stripAnsi` on every
+ * poll tick. (Issue #1897)
+ */
+const COPILOT_STATUS_BAR_SCAN_ROWS = 8;
+// ============================================================================
+// extractResponse (internal)
+// ============================================================================
+/**
+ * Where the tool's pinned footer starts in the capture — the row the transcript
+ * stops at — or -1 when the tool pins none or its landmark is not on the frame.
+ *
+ * Split out of {@link extractResponse} (Issue #3213). The comment above
+ * `openCodeCleanLines` there introduces the first three readers and says why
+ * they stay separate functions; the one in the body here continues it.
+ *
+ * @param cliToolId - CLI tool identifier
+ * @param lines - The trimmed tmux buffer lines array
+ * @param openCodeCleanLines - `lines` with ANSI stripped for opencode, null for every other tool
+ * @returns Index of the first chrome row, or -1
+ */
+function findChromeStart(cliToolId, lines, openCodeCleanLines) {
+    // Issue #2250: Command Code is the fourth. Its landmark is its own — the
+    // `❯ Ask your question...` composer fenced by two full-pane rules, with the
+    // permission-mode row underneath — and it is load-bearing rather than tidy:
+    // that placeholder is drawn with the same `❯ <text>` shape as a transcript
+    // echo, so without the boundary `findRecentUserPromptIndex` anchors the turn
+    // on the FOOTER and every reply extracts as empty (#1289's defect, verbatim).
+    //
+    // Issue #2400: codex is the fifth, and the one that had been missing. It pins
+    // the same two rows — `› Ask Codex to do anything` and the `model · cwd`
+    // status bar — and without a boundary the saturated-window anchor (#1670)
+    // walked into them: the newest `›` in the pane was the COMPOSER, so extraction
+    // started on the status bar and every reply on a saturated pane was saved as
+    // that one row. `findCodexChromeStart` reads the composer by its SGR
+    // attributes (#2310) rather than by its placeholder wording, which is what the
+    // previous guard did and why it stopped working at codex 0.15x.
+    return cliToolId === 'claude'
+        ? (0, cli_patterns_1.findClaudeChromeStart)(lines)
+        : cliToolId === 'copilot'
+            ? (0, cli_patterns_1.findCopilotChromeStart)(lines)
+            : cliToolId === 'command-code'
+                ? (0, cli_patterns_1.findCommandCodeChromeStart)(lines)
+                : cliToolId === 'codex'
+                    ? (0, cli_patterns_1.findCodexChromeStart)(lines)
+                    : openCodeCleanLines
+                        ? (0, cli_patterns_1.findOpenCodeChromeStart)(openCodeCleanLines)
+                        : -1;
+}
+/**
+ * Index of the newest echoed user prompt in the transcript, or -1 when there is
+ * none in the window.
+ *
+ * The body of the `findRecentUserPromptIndex` closure in {@link extractResponse},
+ * split out as it was (Issue #3213). The closure is still there: it binds the
+ * frame and keeps the default window, because `resolveExtractionStartIndex` and
+ * `buildPromptExtractionResult` take the search as a one-argument callback.
+ *
+ * @param cliToolId - CLI tool identifier
+ * @param lines - The trimmed tmux buffer lines array
+ * @param openCodeCleanLines - `lines` with ANSI stripped for opencode, null for every other tool
+ * @param chromeStart - What `findChromeStart` returned for this frame
+ * @param contentEnd - Where the transcript stops: `chromeStart`, or the line count when that is -1
+ * @param windowSize - How many rows up from `contentEnd` to search
+ * @returns Index into `lines`, or -1
+ */
+function findRecentUserPromptIndexInFrame(cliToolId, lines, openCodeCleanLines, chromeStart, contentEnd, windowSize) {
+    let userPromptPattern;
+    if (cliToolId === 'codex') {
+        // Issue #2400: codex's three uses of `›` are separated by their SGR
+        // attributes, not by their text (#2310). This branch used to exclude the
+        // composer with a negative lookahead over its placeholder strings
+        // (`Implement`, `Find and fix`, `Type`, `Summarize`) — codex 0.1x wording,
+        // none of which 0.15x draws. `Ask Codex to do anything` passed the guard,
+        // became the newest "echo", and on a saturated pane (#1670) — the only
+        // path where this anchor decides where extraction STARTS — every reply was
+        // saved as the single status-bar row below it.
+        //
+        // Two independent things now keep the composer out, and the reader needs
+        // both because neither covers the other's frames: `contentEnd` cuts the
+        // composer off structurally when `findCodexChromeStart` located it, and
+        // when it did not, `findCodexUserEchoIndex` steps over the bottom-most
+        // `›` row instead. The second is what still answers on an ANSI-stripped
+        // capture, where none of #2310's attributes survive to be read.
+        return (0, cli_patterns_1.findCodexUserEchoIndex)(lines, contentEnd, windowSize, chromeStart >= 0);
+    }
+    else if (openCodeCleanLines) {
+        // Issue #1911: anchor on the newest ECHOED USER PROMPT, not on the
+        // second-to-last `▣ Build` row. The old anchor belonged to the PREVIOUS
+        // turn, so the echoed prompt of the current one was always extracted as
+        // part of the reply — and on the first turn of a session, where there is
+        // no second marker, it fell through to line 0 and the whole pane (banner
+        // included) became the answer. `windowSize` is ignored: the alternate
+        // screen has no scrollback, so the whole pane IS the window, and every
+        // caller already passes `totalLines` or more for this tool.
+        return (0, response_extractor_1.resolveOpenCodeTurnRegion)(openCodeCleanLines).echoEnd;
+    }
+    else if (cliToolId === 'copilot') {
+        // Issue #1897: copilot 1.0.80 draws the transcript one column in, so the
+        // bare `^[>❯]` form below never matched the echoed prompt -- every copilot
+        // extraction fell back to line 0, i.e. to the launch banner. The composer,
+        // which IS at column 0, lives below `contentEnd` and so cannot be picked up
+        // as an echo here.
+        //
+        // The scan then walks past the echo's own wrapped rows and returns the LAST
+        // of them, so that callers' `+ 1` lands on the reply rather than on the
+        // second half of the operator's question.
+        //
+        // Same defect as #1911's opencode branch above and the same shape of fix,
+        // but NOT the same code: opencode's echo is a `┃  <text>` gutter row and
+        // copilot's is ` ❯ <text>` at the pane's one-column indent, so the anchor
+        // and the continuation rule are both tool-specific measurements.
+        for (let i = contentEnd - 1; i >= Math.max(0, contentEnd - windowSize); i--) {
+            if (!cli_patterns_1.COPILOT_USER_ECHO_PATTERN.test((0, cli_patterns_1.stripAnsi)(lines[i])))
+                continue;
+            let echoEnd = i;
+            while (echoEnd + 1 < contentEnd &&
+                cli_patterns_1.COPILOT_TRANSCRIPT_CONTINUATION_PATTERN.test((0, cli_patterns_1.stripAnsi)(lines[echoEnd + 1]))) {
+                echoEnd++;
+            }
+            return echoEnd;
+        }
+        return -1;
+    }
+    else {
+        userPromptPattern = /^[>❯]\s+\S/;
+    }
+    // Issue #1289: for Claude the search stops above the footer. The text the
+    // user just typed sits in the footer's input box and matches the same "❯ …"
+    // shape as the transcript echo; anchoring on it would treat the footer as
+    // the newest turn and extract the status bar as its reply.
+    for (let i = contentEnd - 1; i >= Math.max(0, contentEnd - windowSize); i--) {
+        const cleanLine = (0, cli_patterns_1.stripAnsi)(lines[i]);
+        if (userPromptPattern.test(cleanLine)) {
+            return i;
+        }
+    }
+    return -1;
+}
+// Startup-screen defenses, split out of extractResponse one function per tool
+// (Issue #3213). Each is called on a frame the completion rules have already
+// accepted, with the `response` extracted from it, and answers only for its own
+// tool: the incomplete result to return when the frame is that tool's startup
+// screen rather than a reply, `null` when the response stands. They are declared
+// in the order extractResponse calls them — the comments inside say "above" and
+// "below" about each other and about the checks that ran before the call.
+//
+// Issue #3293 added the last two, codex and vibe-local. Theirs is the one result
+// here that is not incomplete: an empty COMPLETE one, for the reason given at
+// `readStartupScreenPastCursor`.
+/**
+ * Claude: the startup banner, and an echoed prompt with no reply under it yet.
+ *
+ * @param cliToolId - CLI tool identifier
+ * @param response - The response extracted from the frame
+ * @param totalLines - Total line count in the buffer
+ * @param skipPatterns - The tool's skip patterns from `getCliToolPatterns`
+ * @param findRecentUserPromptIndex - Callback to locate the most recent user prompt
+ * @returns The incomplete result to return, or null when the response stands
+ */
+function suppressClaudeStartupScreen(cliToolId, response, totalLines, skipPatterns, findRecentUserPromptIndex) {
+    // CRITICAL FIX: Detect and skip Claude Code startup banner/screen
+    if (cliToolId === 'claude') {
+        const cleanResponse = (0, cli_patterns_1.stripAnsi)(response);
+        // Issue #2247: `│` is what Claude Code draws markdown TABLES with -- the
+        // live frame in `tests/fixtures/claude-live-2247/turn-table.txt` is a
+        // two-row table and nothing else -- so it identified a reply, not a banner.
+        // The banner's own frame glyphs are the rounded corners and the block
+        // shading; those stay.
+        const hasBannerArt = /[╭╮╰╯]/.test(cleanResponse) || /░{3,}/.test(cleanResponse) || /▓{3,}/.test(cleanResponse);
+        // Issue #2247: the bare `v\d+\.\d+` alternative matched any version string a
+        // reply happens to mention. The frame that lost a turn on 2026-09-02 was
+        // "GitHub Release v0.30.0 を公開しました" (148 chars, well under the 2000
+        // below). What the banner actually prints is the tool's own name and
+        // version on one row -- `Claude Code v2.1.258` -- so that is what is
+        // matched now, plus the `claude/` form older banners used.
+        const hasVersionInfo = /Claude Code v\d+\.\d+|claude\//.test(cleanResponse);
+        const hasStartupTips = /Tip:|for shortcuts|\?\s*for help/.test(cleanResponse);
+        const hasProjectInit = /^\s*\/Users\/.*$/m.test(cleanResponse) && cleanResponse.split('\n').length < 30;
+        // Issue #2247: the anchors above are only evidence of a banner on a pane
+        // that has not had a single turn yet -- the same shape as the #1897 copilot
+        // fix below. Claude echoes every prompt into the transcript as `❯ <text>`,
+        // and the startup screen has none, so an echo anywhere in the transcript
+        // rules the banner out no matter what the reply quotes.
+        //
+        // The search is `findRecentUserPromptIndex`, deliberately: it is the same
+        // `/^[>❯]\s+\S/` this file already anchors extraction on, and it stops at
+        // `contentEnd`. That bound is load-bearing rather than incidental -- the
+        // footer's composer draws a DIM ghost suggestion (`❯ Try "write a test for
+        // <filepath>"`, see `boot-banner.txt`) whose stripped bytes are identical
+        // to a real echo (#1879), so a scan over the whole pane would read the
+        // startup screen as "already had a turn" and put the banner back in
+        // History.
+        const hasTurnEcho = findRecentUserPromptIndex(totalLines) >= 0;
+        const userPromptMatch = cleanResponse.match(/^[>❯]\s+(\S.*)$/m);
+        if (userPromptMatch) {
+            const userPromptIndex = cleanResponse.indexOf(userPromptMatch[0]);
+            const contentAfterPrompt = cleanResponse.substring(userPromptIndex + userPromptMatch[0].length).trim();
+            const contentLines = contentAfterPrompt.split('\n').filter(line => {
+                const trimmed = line.trim();
+                return trimmed &&
+                    !skipPatterns.some(p => p.test(trimmed)) &&
+                    !/^─+$/.test(trimmed);
+            });
+            if (contentLines.length === 0) {
+                return (0, response_checker_extraction_result_1.incompleteResult)(totalLines);
+            }
+        }
+        else if (!hasTurnEcho &&
+            (hasBannerArt || hasVersionInfo || hasStartupTips || hasProjectInit) &&
+            response.length < 2000) {
+            // Issue #2247: this branch used to swallow the turn in silence -- the
+            // poller kept ticking every 2s and `response-poller` logged nothing at
+            // all, so the only way to tell a lost turn from an idle session was to
+            // re-run `extractResponse` on a saved pane by hand. It is reached only
+            // before the first echo lands, so it cannot become a per-tick flood.
+            logger.info('Claude startup banner suppressed, response not saved', {
+                responseLength: response.length,
+                hasBannerArt,
+                hasVersionInfo,
+                hasStartupTips,
+                hasProjectInit,
+            });
+            return (0, response_checker_extraction_result_1.incompleteResult)(totalLines);
+        }
+    }
+    return null;
+}
+/**
+ * Copilot: the launch screen.
+ *
+ * @param cliToolId - CLI tool identifier
+ * @param response - The response extracted from the frame
+ * @param totalLines - Total line count in the buffer
+ * @returns The incomplete result to return, or null when the response stands
+ */
+function suppressCopilotLaunchScreen(cliToolId, response, totalLines) {
+    // Issue #1897: copilot's launch screen is a complete, idle frame -- composer
+    // drawn, key hints on the status bar -- so every check above accepts it and
+    // History used to open with the banner ("Current Sessions Issues Pull
+    // requests Gists / No copilot-instructions.md found… / Tip: /app") saved as
+    // the agent's first reply, before the operator had said anything.
+    //
+    // What actually distinguishes it is that no turn has happened: copilot echoes
+    // every prompt into the transcript as ` ❯ <text>`, and the launch screen has
+    // none. The banner anchors are only consulted once that echo is missing, so a
+    // reply that quotes any of this wording is unaffected.
+    if (cliToolId === 'copilot') {
+        const cleanResponse = (0, cli_patterns_1.stripAnsi)(response);
+        const hasUserEcho = cleanResponse
+            .split('\n')
+            .some(line => cli_patterns_1.COPILOT_USER_ECHO_PATTERN.test(line));
+        if (!hasUserEcho && cli_patterns_1.COPILOT_BOOT_BANNER_ANCHORS.some(anchor => anchor.test(cleanResponse))) {
+            return (0, response_checker_extraction_result_1.incompleteResult)(totalLines);
+        }
+    }
+    return null;
+}
+/**
+ * Command Code: the launch screen.
+ *
+ * @param cliToolId - CLI tool identifier
+ * @param response - The response extracted from the frame
+ * @param totalLines - Total line count in the buffer
+ * @param findRecentUserPromptIndex - Callback to locate the most recent user prompt
+ * @returns The incomplete result to return, or null when the response stands
+ */
+function suppressCommandCodeLaunchScreen(cliToolId, response, totalLines, findRecentUserPromptIndex) {
+    // Issue #2250: Command Code's launch screen is a complete, idle frame --
+    // block-art logo, three `#` banner rows, composer drawn between its two
+    // rules -- so every check above accepts it, and History would open with
+    // `# Command Code v1.40.1 / # models: … / # <cwd>` saved as the agent's
+    // first reply before the operator has said anything.
+    //
+    // The rule is ONE condition and it is a positive one: Command Code echoes
+    // every prompt into the transcript as `❯ <text>`, and the launch screen has
+    // none. That is the shape #1897 and #2247 both converged on; the anchor
+    // heuristics claude carries above (`hasBannerArt` / `hasVersionInfo` /
+    // `hasStartupTips`) are deliberately NOT reproduced here, because they are
+    // what #2247 had to take back -- a bare version string in a reply is a
+    // normal reply, and Command Code prints its own version on every launch.
+    //
+    // `findRecentUserPromptIndex` is the same `/^[>❯]\s+\S/` scan the extraction
+    // anchors on, and it stops at `contentEnd`, so the composer's own dim
+    // `❯ Ask your question...` placeholder cannot be mistaken for an echo
+    // (#1879's trap).
+    if (cliToolId === 'command-code' && findRecentUserPromptIndex(totalLines) < 0) {
+        logger.info('Command Code launch screen suppressed, response not saved', {
+            responseLength: response.length,
+        });
+        return (0, response_checker_extraction_result_1.incompleteResult)(totalLines);
+    }
+    return null;
+}
+/**
+ * Gemini: the banner, the auth/loading states, and a frame too short to be a reply.
+ *
+ * @param cliToolId - CLI tool identifier
+ * @param response - The response extracted from the frame
+ * @param totalLines - Total line count in the buffer
+ * @returns The incomplete result to return, or null when the response stands
+ */
+function suppressGeminiStartupScreen(cliToolId, response, totalLines) {
+    // Gemini-specific check
+    if (cliToolId === 'gemini') {
+        const bannerCharCount = (response.match(/[░█]/g) || []).length;
+        const totalChars = response.length;
+        if (bannerCharCount > totalChars * 0.3) {
+            return (0, response_checker_extraction_result_1.incompleteResult)(totalLines);
+        }
+        if (response_poller_constants_1.GEMINI_LOADING_INDICATORS.some(indicator => response.includes(indicator))) {
+            return (0, response_checker_extraction_result_1.incompleteResult)(totalLines);
+        }
+        if (!response.includes('\u2726') && response.length < 10) {
+            return (0, response_checker_extraction_result_1.incompleteResult)(totalLines);
+        }
+    }
+    return null;
+}
+/**
+ * The result for a startup screen of a tool that reads its reply from the
+ * cursor: nothing to save, and the cursor moved past the screen (Issue #3293).
+ *
+ * The defenses above return an incomplete result and leave the cursor where it
+ * was. Where extraction is anchored on the echoed prompt (copilot,
+ * command-code, opencode) that costs nothing: once a turn exists, the banner
+ * above its echo is never read. codex and vibe-local start at
+ * `lastCapturedLine`. A cursor left at 0 would put the banner back on the first
+ * turn — vibe-local's first reply would be saved with the whole startup screen
+ * above it.
+ *
+ * So the screen is reported as read: a COMPLETE result with an empty response
+ * and the `lineCount` the banner save had. `checkForResponse` writes that
+ * cursor and saves nothing ("Validate response content is not empty"), which
+ * leaves every later tick with exactly the cursor it had before this Issue.
+ *
+ * Asked of the whole pane, by the reader the pre-send flush uses for the same
+ * question ({@link isStartupScreenWithoutUserEcho}), and not of a clipped
+ * capture: the echo of a turn longer than the window has scrolled out (#1670).
+ *
+ * @param tool - The tool the calling defense answers for
+ * @param ctx - What this call has read off the capture
+ * @param response - The response extracted from the frame
+ * @param endIndex - The line count `collectCompletedResponse` reported for it
+ * @returns The empty complete result to return, or null when the response stands
+ */
+function readStartupScreenPastCursor(tool, ctx, response, endIndex) {
+    const { cliToolId, lines, lastCapturedLine, bufferReset, captureWindowSaturated } = ctx;
+    // An empty response is already what this would return.
+    if (cliToolId !== tool || !response || captureWindowSaturated) {
+        return null;
+    }
+    if (!(0, startup_screen_1.isStartupScreenWithoutUserEcho)(cliToolId, lines)) {
+        return null;
+    }
+    // Only when `checkForResponse` goes on to write the cursor, which is once per
+    // startup screen. vibe-local's cursor ends on the last row, so every later
+    // tick of the idle pane takes the buffer-reset branch, re-extracts the banner
+    // and arrives here again.
+    if (bufferReset || endIndex > lastCapturedLine) {
+        logger.info(`${tool} startup screen suppressed, response not saved`, {
+            responseLength: response.length,
+            lineCount: endIndex,
+        });
+    }
+    return {
+        response: '',
+        isComplete: true,
+        lineCount: endIndex,
+        bufferReset,
+        captureWindowSaturated,
+    };
+}
+/**
+ * Codex: the startup screen.
+ *
+ * @param ctx - What this call has read off the capture
+ * @param response - The response extracted from the frame
+ * @param endIndex - The line count `collectCompletedResponse` reported for it
+ * @returns The empty complete result to return, or null when the response stands
+ */
+function suppressCodexStartupScreen(ctx, response, endIndex) {
+    // Issue #3293: codex's startup screen is a complete, idle frame — composer
+    // drawn, no status line above it — so `hasPrompt && !isThinking` accepts it,
+    // and the rows from the cursor down were saved as the agent's first reply.
+    // On 0.160.0 that is `>_ OpenAI Codex (v0.160.0)`, the cwd, a tagline and the
+    // block-art logo (20 rows in `tests/fixtures/startup-screen-3293/`); on 0.15x
+    // the `Tip:` and usage notices under the banner box, whose own rows the skip
+    // patterns happen to drop.
+    //
+    // It took one poll of that screen. The frame the folder-trust dialog leaves
+    // behind when it is answered is this screen, and answering it with `respond`
+    // starts the poller.
+    //
+    // The dialog itself never reaches this point: it has no composer, and it was
+    // returned as a prompt by the early check in extractResponse.
+    //
+    // A turn longer than the pane does reach it, and has to pass. 0.160.0 keeps
+    // no scrollback, so the echo of such a turn has left the pane off the top and
+    // "composer, no echo above it" holds for a finished reply. The reader asks
+    // for the banner row as well, which that pane has lost with the echo.
+    //
+    // Issue #3335: on 0.160.0 the cursor this leaves (the composer row, 996) is
+    // also where it stays — the pane does not grow, so the replies of the
+    // session are drawn above it and the screen read takes none of them. That
+    // is by design: they reach History from codex's transcript. See the codex
+    // branch of `resolveExtractionStartIndex`.
+    return readStartupScreenPastCursor('codex', ctx, response, endIndex);
+}
+/**
+ * Vibe Local: the startup screen.
+ *
+ * @param ctx - What this call has read off the capture
+ * @param response - The response extracted from the frame
+ * @param endIndex - The line count `collectCompletedResponse` reported for it
+ * @returns The empty complete result to return, or null when the response stands
+ */
+function suppressVibeLocalStartupScreen(ctx, response, endIndex) {
+    // Issue #3293: vibe-local's startup screen ends in an empty `ctx:N% ❯` prompt
+    // row with nothing working, so it is accepted the same way. Its skip patterns
+    // name some banner rows (`Model  …`, `CWD  …`) and miss the rest — the
+    // launcher's `Model: …` / `Ollama: …` box, the logo, the hint rows — so most
+    // of the screen was saved (20 rows of the 1.3.3 capture in the same fixtures).
+    return readStartupScreenPastCursor('vibe-local', ctx, response, endIndex);
+}
+/**
+ * Does this frame show a finished turn, by its tool's own completion rule?
+ *
+ * The readings and the four per-tool rules that decided the completion branch
+ * of {@link extractResponse}, split out as they were (Issue #3213). The return
+ * is the condition that `if` tested.
+ *
+ * @param ctx - What this call has read off the capture
+ * @returns True when the turn on the frame is finished
+ */
+function isTurnComplete(ctx) {
+    const { cliToolId, lines, totalLines, checkLineCount, cleanOutputToCheck, promptPattern, separatorPattern, thinkingPattern, } = ctx;
+    const hasPrompt = promptPattern.test(cleanOutputToCheck);
+    const hasSeparator = separatorPattern.test(cleanOutputToCheck);
+    // Issue #1671: Codex's activity markers are past-tense transcript records that
+    // never leave the scrollback, so testing them against this fixed tail window
+    // reports "still thinking" for a finished turn whenever its final message was
+    // short enough to keep the last "• Ran <cmd>" row inside the window. Codex gets
+    // a liveness check that keys off the status line it repaints above the composer
+    // instead; every other tool keeps the tail-window match.
+    const isThinking = cliToolId === 'codex'
+        ? (0, cli_patterns_1.isCodexTurnActive)(lines, checkLineCount)
+        : thinkingPattern.test(cleanOutputToCheck);
+    // Issue #1897: copilot's `hasPrompt` is worthless as a completion signal and
+    // its `isThinking` is worthless as a liveness one. The `❯` composer is drawn
+    // between its two rules throughout a turn (measured on every frame of #1885's
+    // running fixtures), and `COPILOT_THINKING_PATTERN` matches nothing copilot
+    // 1.0.80 draws (0 of 44 live generating frames). So `hasPrompt && !isThinking`
+    // was true on the very first poll of a running turn -- the extractor declared
+    // the turn finished, saved the status bar as the reply, and `checkForResponse`
+    // stopped polling, which is why the real answer never reached History.
+    //
+    // 1.0.80 paints the turn's state on the bottom row of the pane and nowhere
+    // else, so that ROW -- never a tail window, which copilot's own reply text can
+    // forge (`status-vocabulary-in-response.txt`) -- is the evidence. `idle` is a
+    // positive observation that the turn is over (design policy §4 D1 decision 1
+    // item 2); `working` and `null` (a dialog box has taken the bar away) both mean
+    // "not finished", and the dialog case is already served by the prompt path
+    // above.
+    const copilotStatusBar = cliToolId === 'copilot'
+        ? (0, cli_patterns_1.readCopilotStatusBar)(lines.slice(Math.max(0, totalLines - COPILOT_STATUS_BAR_SCAN_ROWS)).map(cli_patterns_1.stripAnsi))
+        : null;
+    // Prompt-based completion logic
+    const isPromptBasedComplete = cliToolId === 'copilot'
+        ? copilotStatusBar === 'idle'
+        : (cliToolId === 'codex' || cliToolId === 'gemini' || cliToolId === 'vibe-local' || cliToolId === 'antigravity') && hasPrompt && !isThinking;
+    const isClaudeComplete = cliToolId === 'claude' && hasPrompt && hasSeparator && !isThinking;
+    // Issue #2250: claude's shape, because Command Code's layout is claude's — the
+    // composer sits between two full-pane rules and is drawn only when the agent
+    // will accept input. Deliberately NOT keyed on `✻ Worked for`: that row is the
+    // live turn's, not the transcript's (it is present in `turn-version.txt` and
+    // gone from `dialog-create-file.txt`, the same pane one prompt later) and
+    // `WorkedDurationNote` omits it entirely for a turn under 1000 ms.
+    const isCommandCodeComplete = cliToolId === 'command-code' && hasPrompt && hasSeparator && !isThinking;
+    const isOpenCodeDone = cliToolId === 'opencode' && (0, response_extractor_1.isOpenCodeComplete)(cleanOutputToCheck);
+    return isPromptBasedComplete || isClaudeComplete || isCommandCodeComplete || isOpenCodeDone;
+}
+/**
+ * Collect the rows of a finished turn's reply, and where the cursor stops.
+ *
+ * The first half of the completion branch of {@link extractResponse}, split out
+ * as it was (Issue #3213). The loop writes `endIndex` when it stops early, so
+ * it is returned beside the response instead of being written through.
+ *
+ * @param ctx - What this call has read off the capture
+ * @returns The trimmed response, and the line count to report for it
+ */
+function collectCompletedResponse(ctx) {
+    const { cliToolId, lines, totalLines, chromeStart, contentEnd, lastCapturedLine, bufferReset, captureWindowSaturated, skipPatterns, findRecentUserPromptIndex, } = ctx;
+    const responseLines = [];
+    const startIndex = (0, response_extractor_1.resolveExtractionStartIndex)(lastCapturedLine, totalLines, bufferReset, cliToolId, findRecentUserPromptIndex, captureWindowSaturated);
+    // `contentEnd` bounds the content only; `endIndex` keeps reporting the full
+    // buffer so lineCount bookkeeping in session_states is unchanged (#1289).
+    //
+    // Issue #2400: codex is the exception, and it is the pre-existing behaviour
+    // rather than a new rule. Before this Issue the loop below stopped on the
+    // composer's `›` and wrote that row's index into `endIndex`; now the composer
+    // is outside `contentEnd`, so the break can no longer fire on it and
+    // `endIndex` would silently advance ~3 rows further. Those rows matter for
+    // codex specifically: it renders INLINE, and it repaints the composer band in
+    // place — the next turn's transcript is printed over exactly the rows the
+    // composer occupied in this capture. A cursor parked past them would skip
+    // real content on the following poll. So the cursor stops where the content
+    // stops, which is what it did before.
+    let endIndex = cliToolId === 'codex' ? contentEnd : totalLines;
+    for (let i = startIndex; i < contentEnd; i++) {
+        const line = lines[i];
+        const cleanLine = (0, cli_patterns_1.stripAnsi)(line);
+        if (cliToolId === 'codex' && /^›\s+/.test(cleanLine)) {
+            endIndex = i;
+            break;
+        }
+        if (cliToolId === 'gemini' && /^(%|\$|.*@.*[%$#])\s*$/.test(cleanLine)) {
+            endIndex = i;
+            break;
+        }
+        // Antigravity (agy): the bare ">" input box line marks the end of the
+        // response (the status bar and shortcuts footer follow below it). (Issue #988)
+        if (cliToolId === 'antigravity' && /^>\s*$/.test(cleanLine)) {
+            endIndex = i;
+            break;
+        }
+        // Issue #1911: both rows this stops on (`Ask anything...` in the composer,
+        // `tab agents  ctrl+p commands` under its border) live in the chrome, which
+        // `contentEnd` now excludes structurally. Kept only as the fallback for a
+        // frame whose chrome could not be located, because there it is still the
+        // one boundary available — and #1883 measured that a REPLY can contain
+        // `Ask anything...`, so cutting the turn on it is a last resort, not the
+        // primary rule.
+        if (cliToolId === 'opencode' && chromeStart < 0) {
+            if (cli_patterns_1.OPENCODE_PROMPT_PATTERN.test(cleanLine) || cli_patterns_1.OPENCODE_PROMPT_AFTER_RESPONSE.test(cleanLine)) {
+                endIndex = i;
+                break;
+            }
+        }
+        const shouldSkip = skipPatterns.some(pattern => pattern.test(cleanLine));
+        if (shouldSkip) {
+            continue;
+        }
+        responseLines.push(line);
+    }
+    const response = responseLines.join('\n').trim();
+    return { response, endIndex };
+}
+/**
+ * Build the result for a frame the completion rules accepted.
+ *
+ * The body of the completion branch of {@link extractResponse}, split out as it
+ * was (Issue #3213). Every return is `extractResponse`'s own result: incomplete
+ * when the reply's tail still shows a thinking indicator or the frame is a
+ * startup screen, complete otherwise. (A codex or vibe-local startup screen is
+ * the exception — complete, with an empty response; Issue #3293.)
+ *
+ * @param ctx - What this call has read off the capture
+ * @returns What `extractResponse` returns for this frame
+ */
+function extractCompletedResponse(ctx) {
+    const { cliToolId, totalLines, openCodeCleanLines, bufferReset, captureWindowSaturated, thinkingPattern, skipPatterns, findRecentUserPromptIndex, } = ctx;
+    const { response, endIndex } = collectCompletedResponse(ctx);
+    // DR-004: Check only the tail of the response for thinking indicators.
+    //
+    // Issue #1897: not for copilot. This is the same tail-window match the #1671
+    // codex fix removed from the liveness test, and on copilot it is both
+    // redundant and harmful: the status bar above has already made a positive
+    // `idle` observation about THIS frame, while the window here sees transcript
+    // that never scrolls away. `COPILOT_THINKING_PATTERN`'s braille alternative
+    // matches any spinner glyph a reply happens to quote, and the turn would then
+    // be reported unfinished for the rest of the session.
+    const responseTailLines = response.split('\n').slice(-thinking_constants_1.THINKING_TAIL_LINE_COUNT).join('\n');
+    if (cliToolId !== 'copilot' && thinkingPattern.test(responseTailLines)) {
+        return (0, response_checker_extraction_result_1.incompleteResult)(totalLines);
+    }
+    // Startup-screen defenses, one per tool and in this order. Each answers only
+    // for its own tool; `null` means the response stands.
+    const startupScreen = suppressClaudeStartupScreen(cliToolId, response, totalLines, skipPatterns, findRecentUserPromptIndex) ??
+        suppressCopilotLaunchScreen(cliToolId, response, totalLines) ??
+        suppressCommandCodeLaunchScreen(cliToolId, response, totalLines, findRecentUserPromptIndex) ??
+        suppressGeminiStartupScreen(cliToolId, response, totalLines) ??
+        suppressCodexStartupScreen(ctx, response, endIndex) ??
+        suppressVibeLocalStartupScreen(ctx, response, endIndex);
+    if (startupScreen) {
+        return startupScreen;
+    }
+    return {
+        response,
+        isComplete: true,
+        lineCount: endIndex,
+        bufferReset,
+        captureWindowSaturated,
+        // Issue #1911: opencode only. `echoEnd < 0` means the turn is longer than
+        // the alternate-screen pane and its head has already scrolled away, so
+        // `response` starts mid-answer. Nothing else in this frame can recover it.
+        turnHeadTruncated: openCodeCleanLines
+            ? (0, response_extractor_1.resolveOpenCodeTurnRegion)(openCodeCleanLines).headTruncated
+            : undefined,
+    };
+}
+/**
+ * Build the result for a frame whose turn is still running: what has streamed
+ * so far, or an empty incomplete result when nothing has.
+ *
+ * The tail of {@link extractResponse}, split out as it was (Issue #3213). Both
+ * returns are `extractResponse`'s own result.
+ *
+ * @param ctx - What this call has read off the capture
+ * @returns What `extractResponse` returns for this frame
+ */
+function extractPartialResponse(ctx) {
+    const { lines, totalLines, contentEnd, lastCapturedLine, bufferReset, captureWindowSaturated, skipPatterns, findRecentUserPromptIndex, } = ctx;
+    // Partial response in progress
+    const responseLines = [];
+    const endIndex = totalLines;
+    // Issue #1670: a saturated window makes lastCapturedLine meaningless here too —
+    // starting the partial slice at it would stream an arbitrary tail fragment
+    // instead of the turn so far. Re-anchor on the echoed user prompt.
+    const partialBufferReset = bufferReset || captureWindowSaturated || lastCapturedLine >= endIndex - 5;
+    const recentPromptIndex = partialBufferReset ? findRecentUserPromptIndex(80) : -1;
+    const startIndex = partialBufferReset
+        ? (recentPromptIndex >= 0 ? recentPromptIndex + 1 : Math.max(0, endIndex - 80))
+        : Math.max(0, lastCapturedLine);
+    // Partial (still-streaming) content is bounded by the footer too (#1289).
+    for (let i = startIndex; i < Math.min(endIndex, contentEnd); i++) {
+        const line = lines[i];
+        const cleanLine = (0, cli_patterns_1.stripAnsi)(line);
+        const shouldSkip = skipPatterns.some(pattern => pattern.test(cleanLine));
+        if (shouldSkip) {
+            continue;
+        }
+        responseLines.push(line);
+    }
+    const partialResponse = responseLines.join('\n').trim();
+    if (partialResponse) {
+        return {
+            response: partialResponse,
+            isComplete: false,
+            lineCount: endIndex,
+        };
+    }
+    // Response not yet complete
+    return (0, response_checker_extraction_result_1.incompleteResult)(totalLines);
+}