app/src/utils/agentMessageBubbles.ts - main - GreyCat/openhuman

app/src/utils/agentMessageBubbles.ts
269 строк9 KB
ly-wang19
fix(chat): don't fold a pipe-bearing prose line into a preceding table (#3978)
23 июн 2026, 21:47
23 июн 2026, 21:47e15f90e
import debugFactory from 'debug';

const debug = debugFactory('agent-message-bubbles');

/**
 * Split an agent message into render-time bubble segments.
 *
 * Normalize excessive vertical whitespace first, then split only on double
 * newlines. Fenced code blocks stay intact as a single segment so
 * Markdown/code rendering does not fragment unexpectedly.
 * Markdown tables also stay grouped so they can render as dedicated table UI.
 *
 * Finally, regroup so a section heading and the body that follows it always
 * share one bubble (issue #3807): structured output such as the morning
 * briefing writes `## Heading\n\n- body…`, and splitting on the blank line
 * would otherwise orphan every heading into its own content-less bubble.
 */
export function splitAgentMessageIntoBubbles(content: string): string[] {
  const normalized = content
    .replace(/\r\n/g, '\n')
    .replace(/\n{3,}/g, '\n\n')
    .replace(/\n?\s*<hr\s*\/?>\s*\n?/gi, '\n\n');
  const trimmedContent = normalized.trim();
  if (trimmedContent.length === 0) return [];
  if (!normalized.includes('\n')) {
    return isVisualSeparatorOnly(trimmedContent) ? [] : [trimmedContent];
  }

  const lines = normalized.split('\n');
  const segments: string[] = [];
  let currentLines: string[] = [];
  let inFence = false;

  const flushCurrent = () => {
    const segment = currentLines.join('\n').trim();
    if (segment.length > 0) {
      segments.push(segment);
    }
    currentLines = [];
  };

  for (let index = 0; index < lines.length; index += 1) {
    const line = lines[index];
    const trimmedLine = line.trim();

    if (!inFence && isMarkdownTableStart(lines, index)) {
      if (currentLines.length > 0) {
        flushCurrent();
      }
      // A prose line that merely contains a pipe (e.g. "pick A | B, then run
      // `ps aux | grep x`") satisfies looksLikeMarkdownTableRow but is not a
      // table row. Require the same column count as the header so it isn't
      // absorbed into the table — otherwise parseMarkdownTable rejects the
      // mismatched block (returns null), suppressing the table UI and merging
      // the prose line into the table bubble.
      //
      // Always keep at least the first data row, though: splitMarkdownTableCells
      // splits naively on "|" and does not honor escaped "\|" or pipes inside
      // code spans, so a genuine first row carrying such a pipe over-counts its
      // cells. Dropping it here would leave a header+separator-only block that
      // parseMarkdownTable renders as an empty-row table (and detaches the real
      // row). Keeping it preserves the count mismatch so parseMarkdownTable
      // returns null and the block falls back to readable raw markdown.
      const headerCellCount = splitMarkdownTableCells(line).length;
      const tableLines = [line, lines[index + 1]];
      index += 2;
      let dataRowsAbsorbed = 0;
      while (index < lines.length && looksLikeMarkdownTableRow(lines[index])) {
        const matches = splitMarkdownTableCells(lines[index]).length === headerCellCount;
        if (!matches && dataRowsAbsorbed > 0) break;
        tableLines.push(lines[index]);
        dataRowsAbsorbed += 1;
        index += 1;
      }
      index -= 1;
      segments.push(tableLines.join('\n').trim());
      continue;
    }

    if (trimmedLine.startsWith('```')) {
      if (!inFence && currentLines.length > 0) {
        flushCurrent();
      }
      currentLines.push(line);
      inFence = !inFence;
      if (!inFence) {
        flushCurrent();
      }
      continue;
    }

    if (inFence) {
      currentLines.push(line);
      continue;
    }

    if (trimmedLine.length === 0) {
      if (currentLines.length > 0) {
        flushCurrent();
      }
      continue;
    }

    currentLines.push(line);
  }

  flushCurrent();
  const cleaned = segments.filter(segment => !isVisualSeparatorOnly(segment));
  const grouped = groupSectionsUnderHeadings(cleaned);
  debug(
    '[bubbles] %d char message -> %d raw segment(s) -> %d bubble(s) after heading grouping',
    content.length,
    cleaned.length,
    grouped.length
  );
  return grouped;
}

/**
 * Merge each section heading with the body segments that follow it (up to the
 * next heading) into a single bubble, so a heading is never stranded in a
 * content-less bubble of its own. Segments before the first heading (e.g. a
 * greeting) are left untouched.
 *
 * Tables are deliberately NOT absorbed into a heading bubble: `AgentMessageBubble`
 * only renders the dedicated table UI when the table sits at the start of the
 * bubble (`parseMarkdownTable`), and the Markdown renderer has no GFM-table
 * support, so folding a table behind a heading would render it as raw pipe
 * text. A table therefore stays its own bubble and ends the current section.
 */
function groupSectionsUnderHeadings(segments: string[]): string[] {
  const grouped: string[] = [];
  let index = 0;

  while (index < segments.length) {
    const segment = segments[index];

    if (!startsWithHeading(segment)) {
      grouped.push(segment);
      index += 1;
      continue;
    }

    // Absorb the body paragraphs below this heading — but stop at the next
    // heading or a table, both of which must begin their own bubble.
    const body: string[] = [];
    index += 1;
    while (
      index < segments.length &&
      !startsWithHeading(segments[index]) &&
      !isTableSegment(segments[index])
    ) {
      body.push(segments[index]);
      index += 1;
    }

    if (body.length > 0) {
      grouped.push([segment, ...body].join('\n\n'));
      continue;
    }

    // Heading with no inline body. Avoid an orphan heading bubble: when the
    // next segment is a table the heading labels it from the bubble directly
    // above (and must not be folded into the table); otherwise fold the
    // heading into the previous bubble, unless that bubble is itself a table.
    const nextIsTable = index < segments.length && isTableSegment(segments[index]);
    const previous = grouped[grouped.length - 1];
    if (!nextIsTable && previous !== undefined && !isTableSegment(previous)) {
      grouped[grouped.length - 1] = `${previous}\n\n${segment}`;
    } else {
      grouped.push(segment);
    }
  }

  return grouped;
}

/** A segment "starts a section" when its first non-empty line is a heading. */
function startsWithHeading(segment: string): boolean {
  const firstLine = segment.split('\n').find(line => line.trim().length > 0) ?? '';
  return isHeadingLine(firstLine);
}

/** A segment is a table when its first non-empty line begins a markdown table. */
function isTableSegment(segment: string): boolean {
  const lines = segment.split('\n');
  let index = 0;
  while (index < lines.length && lines[index].trim().length === 0) index += 1;
  return isMarkdownTableStart(lines, index);
}

/**
 * Recognize the heading shapes structured agent output uses to label a
 * section: Markdown ATX headings (`#`..`######`) and a single fully-bold line
 * (`**Calendar**`), optionally with a trailing colon.
 *
 * The bold form deliberately requires the WHOLE line to be bold — inline
 * emphasis with trailing prose (`**Heads up:** the meeting moved`) is NOT a
 * heading, so ordinary emphasized content is not misclassified. The residual
 * heuristic risk is a standalone fully-bold sentence used as content; in that
 * case it is simply merged with the following segment (kept together, never
 * orphaned), which is harmless for the structured briefing shapes this serves.
 */
function isHeadingLine(line: string): boolean {
  const trimmed = line.trim();
  if (/^#{1,6}\s+\S/.test(trimmed)) return true;
  return /^\*\*[^*]+\*\*:?$/.test(trimmed);
}

export interface ParsedMarkdownTable {
  headers: string[];
  rows: string[][];
}

export function parseMarkdownTable(content: string): ParsedMarkdownTable | null {
  const normalized = content.replace(/\r\n/g, '\n').trim();
  const lines = normalized
    .split('\n')
    .map(line => line.trim())
    .filter(line => line.length > 0);

  if (lines.length < 2) return null;
  if (!isMarkdownTableStart(lines, 0)) return null;

  const headers = splitMarkdownTableCells(lines[0]);
  const rows = lines.slice(2).map(splitMarkdownTableCells);

  if (headers.length === 0 || rows.some(row => row.length !== headers.length)) {
    return null;
  }

  return { headers, rows };
}

function isMarkdownTableStart(lines: string[], index: number): boolean {
  const header = lines[index];
  const separator = lines[index + 1];
  if (!header || !separator) return false;
  return looksLikeMarkdownTableRow(header) && looksLikeMarkdownTableSeparator(separator);
}

function looksLikeMarkdownTableRow(line: string): boolean {
  const trimmed = line.trim();
  if (!trimmed.includes('|')) return false;
  const cells = splitMarkdownTableCells(trimmed);
  return cells.length >= 2;
}

function looksLikeMarkdownTableSeparator(line: string): boolean {
  const cells = splitMarkdownTableCells(line);
  if (cells.length < 2) return false;
  return cells.every(cell => /^:?-{3,}:?$/.test(cell));
}

function splitMarkdownTableCells(line: string): string[] {
  return line
    .trim()
    .replace(/^\|/, '')
    .replace(/\|$/, '')
    .split('|')
    .map(cell => cell.trim());
}

function isVisualSeparatorOnly(segment: string): boolean {
  const trimmed = segment.trim();
  if (trimmed.length === 0) return true;
  if (/^<hr\s*\/?>$/i.test(trimmed)) return true;
  return /^(?:-{3,}|\*{3,}|_{3,})$/.test(trimmed.replace(/\s+/g, ''));
}