mirror of
https://github.com/jackwener/OpenCLI.git
synced 2026-09-14 18:25:42 +08:00
4fe9a73ebc
* refactor: migrate adapter imports to package exports Replace all relative imports (../../src/registry.js, ../../browser/cdp.js, etc.) with package exports (@jackwener/opencli/registry, @jackwener/opencli/errors, etc.) across all 484 adapter files. This decouples adapter import resolution from directory structure: - User CLIs in ~/.opencli/clis/ resolve via node_modules symlink - Internal adapters resolve via Node.js self-referencing - No more shim files needed for import resolution Changes: - package.json: add sub-path exports for all public modules - clis/**: replace relative imports with @jackwener/opencli/... - discovery.ts: simplify ensureUserCliCompatShims to symlink-only - registry-api.ts: export CommandArgs type - Remove root-level shim directories (browser/, download/, pipeline/) - Remove shim entries from tsconfig.json include and package.json files * test: add regression tests for package exports Prevents regressions like #788/#791 by: 1. Scanning all adapter files for forbidden relative imports (../../src/, ../../browser/, etc.) — fails if any remain 2. Verifying every package.json export maps to an existing source file 18 new test cases. * fix: use junction on Windows + broaden test patterns - discovery.ts: use 'junction' symlink type on Windows (no admin required) - package-exports.test.ts: generalize forbidden patterns to catch any depth of ../ traversal (not just ../../ and ../../../) * fix: update stale vi.mock/importActual paths in adapter tests Test files still used old relative paths for vi.mock() and vi.importActual() calls. Updated 5 test files to use package exports. Also broadened regression test patterns to catch mock/importActual paths. * fix: use rm instead of unlink for symlink cleanup, add warn on failure Addresses review feedback from Astro-Han: - rm() handles both symlinks and stale directories (unlink fails on dirs) - Log a warning when symlink creation fails instead of silent catch * docs: update import examples to use package exports Update all documentation, contributing guides, and skills to use @jackwener/opencli/registry instead of ../../src/registry.js. Without this, users following the docs would write adapters with broken imports since the old shim files are no longer created.
533 lines
18 KiB
TypeScript
533 lines
18 KiB
TypeScript
import { cli, Strategy } from '@jackwener/opencli/registry';
|
|
import type { IPage } from '@jackwener/opencli/types';
|
|
import { AuthRequiredError, EmptyResultError } from '@jackwener/opencli/errors';
|
|
|
|
interface TimelinePost {
|
|
rank?: number;
|
|
id: string;
|
|
author: string;
|
|
author_url: string;
|
|
headline: string;
|
|
text: string;
|
|
posted_at: string;
|
|
reactions: number;
|
|
comments: number;
|
|
url: string;
|
|
}
|
|
|
|
interface ExtractedBatch {
|
|
loginRequired?: boolean;
|
|
posts?: TimelinePost[];
|
|
}
|
|
|
|
function normalizeWhitespace(value: unknown): string {
|
|
return String(value ?? '').replace(/\s+/g, ' ').trim();
|
|
}
|
|
|
|
function parseMetric(value: unknown): number {
|
|
const raw = normalizeWhitespace(value).toLowerCase();
|
|
if (!raw) return 0;
|
|
|
|
const compact = raw.replace(/,/g, '');
|
|
const match = compact.match(/(\d+(?:\.\d+)?)(k|m)?/i);
|
|
if (!match) return 0;
|
|
|
|
const base = Number(match[1]);
|
|
const suffix = (match[2] || '').toLowerCase();
|
|
if (suffix === 'k') return Math.round(base * 1000);
|
|
if (suffix === 'm') return Math.round(base * 1000000);
|
|
return Math.round(base);
|
|
}
|
|
|
|
function buildPostId(post: Partial<TimelinePost>): string {
|
|
const url = normalizeWhitespace(post.url);
|
|
if (url) return url;
|
|
|
|
const author = normalizeWhitespace(post.author);
|
|
const text = normalizeWhitespace(post.text);
|
|
const postedAt = normalizeWhitespace(post.posted_at);
|
|
return `${author}::${postedAt}::${text.slice(0, 120)}`;
|
|
}
|
|
|
|
function mergeTimelinePosts(existing: TimelinePost[], batch: TimelinePost[]): TimelinePost[] {
|
|
const seen = new Set(existing.map(post => post.id));
|
|
const merged = [...existing];
|
|
|
|
for (const rawPost of batch) {
|
|
const post: TimelinePost = {
|
|
id: buildPostId(rawPost),
|
|
author: normalizeWhitespace(rawPost.author),
|
|
author_url: normalizeWhitespace(rawPost.author_url),
|
|
headline: normalizeWhitespace(rawPost.headline),
|
|
text: normalizeWhitespace(rawPost.text),
|
|
posted_at: normalizeWhitespace(rawPost.posted_at),
|
|
reactions: Number(rawPost.reactions) || 0,
|
|
comments: Number(rawPost.comments) || 0,
|
|
url: normalizeWhitespace(rawPost.url),
|
|
};
|
|
|
|
if (!post.author || !post.text) continue;
|
|
if (seen.has(post.id)) continue;
|
|
seen.add(post.id);
|
|
merged.push(post);
|
|
}
|
|
|
|
return merged;
|
|
}
|
|
|
|
async function extractVisiblePosts(page: IPage): Promise<ExtractedBatch> {
|
|
return page.evaluate(`(function () {
|
|
function normalize(value) {
|
|
return String(value || '').replace(/\\s+/g, ' ').trim();
|
|
}
|
|
function textOf(root, selector) {
|
|
var el = root.querySelector(selector);
|
|
return el ? el.textContent : '';
|
|
}
|
|
function hrefOf(root, selector) {
|
|
var el = root.querySelector(selector);
|
|
return el && el.href ? el.href : '';
|
|
}
|
|
function attrOf(root, selector, attr) {
|
|
var el = root.querySelector(selector);
|
|
return el ? el.getAttribute(attr) : '';
|
|
}
|
|
function cleanTimestamp(value) {
|
|
return normalize(String(value || '').replace(/[•.]/g, ' '));
|
|
}
|
|
function parseMetric(value) {
|
|
var raw = normalize(value).toLowerCase();
|
|
var match;
|
|
var base;
|
|
var suffix;
|
|
if (!raw) return 0;
|
|
match = raw.replace(/,/g, '').match(/(\\d+(?:\\.\\d+)?)(k|m)?/i);
|
|
if (!match) return 0;
|
|
base = Number(match[1]);
|
|
suffix = (match[2] || '').toLowerCase();
|
|
if (suffix === 'k') return Math.round(base * 1000);
|
|
if (suffix === 'm') return Math.round(base * 1000000);
|
|
return Math.round(base);
|
|
}
|
|
function splitBlocks(text) {
|
|
var lines = String(text || '').split('\\n');
|
|
var blocks = [];
|
|
var current = [];
|
|
var i;
|
|
var line;
|
|
for (i = 0; i < lines.length; i += 1) {
|
|
line = normalize(lines[i]);
|
|
if (!line) {
|
|
if (current.length) {
|
|
blocks.push(normalize(current.join(' ')));
|
|
current = [];
|
|
}
|
|
continue;
|
|
}
|
|
current.push(line);
|
|
}
|
|
if (current.length) blocks.push(normalize(current.join(' ')));
|
|
return blocks;
|
|
}
|
|
function looksLikeTimestamp(value) {
|
|
var lower = String(value || '').toLowerCase();
|
|
return /^\\d+\\s*(s|m|h|d|w|mo|yr|min)(\\s*[•.])?$/i.test(lower);
|
|
}
|
|
function looksLikeBadge(value) {
|
|
var lower = String(value || '').toLowerCase();
|
|
return String(value || '').indexOf('•') === 0
|
|
|| lower === '1st'
|
|
|| lower === '2nd'
|
|
|| lower === '3rd'
|
|
|| lower === 'degree connection';
|
|
}
|
|
function looksLikeAction(value) {
|
|
return /^(follow|send message|connect|visit my website|view my newsletter|subscribe)$/i.test((value || '').toLowerCase());
|
|
}
|
|
function looksLikeCta(value) {
|
|
return /^(book an appointment|view my services|visit my website|view my newsletter|subscribe|learn more|contact us)$/i.test((value || '').toLowerCase());
|
|
}
|
|
function looksLikeEngagement(value) {
|
|
return /(reactions?|comments?|reposts?)/i.test(String(value || ''));
|
|
}
|
|
function looksLikeFooterAction(value) {
|
|
return /^(like|comment|repost|send|reply|load more comments)$/i.test((value || '').toLowerCase());
|
|
}
|
|
function findActivityUrn(root) {
|
|
var elements = [root].concat(Array.from(root.querySelectorAll('*')));
|
|
var i;
|
|
var j;
|
|
var attrs;
|
|
var value;
|
|
var match;
|
|
for (i = 0; i < elements.length; i += 1) {
|
|
attrs = Array.from(elements[i].attributes || []);
|
|
for (j = 0; j < attrs.length; j += 1) {
|
|
value = String(attrs[j].value || '');
|
|
match = value.match(/urn:li:activity:\\d+/);
|
|
if (match) return match[0];
|
|
}
|
|
}
|
|
return '';
|
|
}
|
|
function parseReactionCount(root, blocks) {
|
|
var direct = textOf(root, '.social-details-social-counts__reactions-count');
|
|
var rootText = String(root.innerText || '');
|
|
var i;
|
|
var value;
|
|
value = rootText.match(/and\\s+(\\d[\\d,]*)\\s+others\\s+reacted/i);
|
|
if (value) return parseMetric(value[1]) + 1;
|
|
value = rootText.match(/and\\s+(\\d[\\d,]*)\\s+others(?!\\s+comments?)(?!\\s+reposts?)/i);
|
|
if (value) return parseMetric(value[1]) + 1;
|
|
value = rootText.match(/(\\d[\\d,]*)\\s+reactions?/i);
|
|
if (value) return parseMetric(value[0]);
|
|
if (direct) return parseMetric(direct);
|
|
for (i = 0; i < blocks.length; i += 1) {
|
|
value = blocks[i];
|
|
if (/and\\s+\\d[\\d,]*\\s+others(?!\\s+comments?)(?!\\s+reposts?)/i.test(value)) {
|
|
return parseMetric(value) + 1;
|
|
}
|
|
if (/reactions?/i.test(value)) return parseMetric(value);
|
|
if (/and\\s+\\d+[\\d,]*\\s+others\\s+reacted/i.test(value)) return parseMetric(value) + 1;
|
|
}
|
|
return 0;
|
|
}
|
|
function parseCommentCount(blocks) {
|
|
var i;
|
|
var text = blocks.join(' ');
|
|
var match = text.match(/(\\d[\\d,]*)\\s+comments?/i);
|
|
if (match) return parseMetric(match[0]);
|
|
for (i = 0; i < blocks.length; i += 1) {
|
|
if (/comments?/i.test(blocks[i])) return parseMetric(blocks[i]);
|
|
}
|
|
return 0;
|
|
}
|
|
function selectProfileLink(root, author) {
|
|
var links = Array.from(root.querySelectorAll('a[href*="/in/"], a[href*="/company/"]'));
|
|
var normalizedAuthor = normalize(author).toLowerCase();
|
|
var i;
|
|
var label;
|
|
for (i = 0; i < links.length; i += 1) {
|
|
label = normalize(links[i].textContent || links[i].getAttribute('aria-label')).toLowerCase();
|
|
if (!links[i].href) continue;
|
|
if (normalizedAuthor && label.indexOf(normalizedAuthor) >= 0) return links[i];
|
|
}
|
|
return links[0] || null;
|
|
}
|
|
function selectProfileUrl(root, author) {
|
|
var link = selectProfileLink(root, author);
|
|
return link && link.href ? link.href : '';
|
|
}
|
|
function parseActorLinkMeta(root, author) {
|
|
var link = selectProfileLink(root, author);
|
|
var text = normalize(link ? link.textContent : '');
|
|
var normalizedAuthor = normalize(author);
|
|
var match;
|
|
var rest;
|
|
var headline = '';
|
|
var postedAt = '';
|
|
if (!text || !normalizedAuthor) return { headline: '', postedAt: '' };
|
|
if (text.indexOf(normalizedAuthor) === 0) {
|
|
rest = normalize(text.slice(normalizedAuthor.length));
|
|
} else {
|
|
rest = text;
|
|
}
|
|
rest = normalize(rest.replace(/^[•·]\\s*(1st|2nd|3rd\\+?|3rd|degree connection)/i, ''));
|
|
match = rest.match(/(\\d+\\s*(?:s|m|h|d|w|mo|yr|min))\\s*[•·]?$/i);
|
|
if (match) {
|
|
postedAt = cleanTimestamp(match[1]);
|
|
headline = normalize(rest.slice(0, rest.length - match[0].length));
|
|
} else {
|
|
headline = rest;
|
|
}
|
|
headline = normalize(headline.replace(/^(book an appointment|view my services|visit my website|view my newsletter)\\s*/i, ''));
|
|
return { headline: headline, postedAt: postedAt };
|
|
}
|
|
function stripBodyTail(value) {
|
|
return normalize(String(value || '')
|
|
.replace(/\\s+\\d[\\d,]*\\s+reactions?[\\s\\S]*$/i, '')
|
|
.replace(/\\s+\\d[\\d,]*\\s+comments?[\\s\\S]*$/i, '')
|
|
.replace(/\\s+[A-Z][^\\n]+\\s+and\\s+\\d[\\d,]*\\s+others\\s+reacted[\\s\\S]*$/i, '')
|
|
.replace(/\\s+Like\\s+Comment\\s+Repost\\s+Send[\\s\\S]*$/i, '')
|
|
.replace(/\\s+Reaction button state:[\\s\\S]*$/i, '')
|
|
.replace(/^\\d+\\s*(?:s|m|h|d|w|mo|yr|min)\\s*[•.]?\\s*Follow\\s+/i, '')
|
|
);
|
|
}
|
|
function parseActorMeta(root) {
|
|
var actorLink = root.querySelector('a[href*="/in/"], a[href*="/company/"]');
|
|
var actorText = normalize(actorLink ? actorLink.textContent : '');
|
|
var author = '';
|
|
var headline = '';
|
|
var postedAt = '';
|
|
var match;
|
|
if (actorText) {
|
|
match = actorText.match(/^(.+?)\\s+[•·]\\s+(1st|2nd|3rd\\+?|3rd|degree connection)(.*)$/i);
|
|
if (match) {
|
|
author = normalize(match[1]);
|
|
actorText = normalize(match[3]);
|
|
}
|
|
}
|
|
match = actorText.match(/(.+?)\\s+(\\d+\\s*(?:s|m|h|d|w|mo|yr|min))\\s*[•·]?$/i);
|
|
if (match) {
|
|
headline = normalize(match[1]);
|
|
postedAt = cleanTimestamp(match[2]);
|
|
} else if (actorText) {
|
|
headline = actorText;
|
|
}
|
|
return {
|
|
author: author,
|
|
headline: headline,
|
|
postedAt: postedAt,
|
|
authorUrl: actorLink && actorLink.href ? actorLink.href : '',
|
|
};
|
|
}
|
|
function extractFromListItem(root) {
|
|
var blocks = splitBlocks(root.innerText || '');
|
|
var filtered = [];
|
|
var i;
|
|
var value;
|
|
var author = '';
|
|
var authorUrl = '';
|
|
var headline = '';
|
|
var postedAt = '';
|
|
var text = '';
|
|
var bodyStart = -1;
|
|
var permalink;
|
|
var url;
|
|
var reactions;
|
|
var comments;
|
|
var endIndex = -1;
|
|
var urn;
|
|
|
|
if (blocks.length < 5) return null;
|
|
if (blocks[0] !== 'Feed post') return null;
|
|
|
|
for (i = 1; i < blocks.length; i += 1) {
|
|
value = blocks[i];
|
|
if (!value) continue;
|
|
if (/commented on this|reposted this|liked this|suggested/i.test(value)) continue;
|
|
filtered.push(value);
|
|
}
|
|
if (filtered.length < 4) return null;
|
|
|
|
for (i = 0; i < filtered.length; i += 1) {
|
|
value = filtered[i];
|
|
if (!author && !looksLikeBadge(value) && !looksLikeAction(value) && !looksLikeTimestamp(value)) {
|
|
author = value;
|
|
continue;
|
|
}
|
|
if (author && !headline && !looksLikeBadge(value) && !looksLikeAction(value) && !looksLikeTimestamp(value) && !looksLikeCta(value)) {
|
|
headline = value;
|
|
continue;
|
|
}
|
|
if (!postedAt && looksLikeTimestamp(value)) {
|
|
postedAt = cleanTimestamp(value);
|
|
continue;
|
|
}
|
|
}
|
|
|
|
if (!author) return null;
|
|
authorUrl = selectProfileUrl(root, author);
|
|
if (!headline || !postedAt) {
|
|
var actorMeta = parseActorLinkMeta(root, author);
|
|
if (!headline && actorMeta.headline) headline = actorMeta.headline;
|
|
if (!postedAt && actorMeta.postedAt) postedAt = actorMeta.postedAt;
|
|
}
|
|
|
|
for (i = 0; i < filtered.length; i += 1) {
|
|
value = filtered[i];
|
|
if (looksLikeAction(value)) {
|
|
bodyStart = i + 1;
|
|
break;
|
|
}
|
|
}
|
|
if (bodyStart < 0 && postedAt) {
|
|
bodyStart = filtered.indexOf(postedAt) + 1;
|
|
}
|
|
if (bodyStart < 0) bodyStart = Math.min(filtered.length, headline ? 2 : 1);
|
|
|
|
for (i = bodyStart; i < filtered.length; i += 1) {
|
|
value = filtered[i];
|
|
if (looksLikeEngagement(value) || looksLikeFooterAction(value)) {
|
|
endIndex = i;
|
|
break;
|
|
}
|
|
}
|
|
if (endIndex < 0) endIndex = filtered.length;
|
|
|
|
text = stripBodyTail(filtered.slice(bodyStart, endIndex).join('\\n\\n'));
|
|
if (!text) return null;
|
|
|
|
permalink = root.querySelector('a[href*="/feed/update/"], a[href*="/posts/"], a[href*="/pulse/"]');
|
|
url = permalink ? permalink.href : '';
|
|
urn = findActivityUrn(root);
|
|
if (!url && urn) url = 'https://www.linkedin.com/feed/update/' + urn + '/';
|
|
reactions = parseReactionCount(root, filtered);
|
|
comments = parseCommentCount(filtered);
|
|
|
|
return {
|
|
id: url || (author + '::' + postedAt + '::' + text.slice(0, 120)),
|
|
author: author,
|
|
author_url: authorUrl,
|
|
headline: headline,
|
|
text: text,
|
|
posted_at: postedAt,
|
|
reactions: reactions,
|
|
comments: comments,
|
|
url: url,
|
|
};
|
|
}
|
|
function commentMetric(root) {
|
|
var links = Array.from(root.querySelectorAll('button, a'));
|
|
var i;
|
|
var label;
|
|
for (i = 0; i < links.length; i += 1) {
|
|
label = normalize(links[i].textContent || links[i].getAttribute('aria-label'));
|
|
if (/comment/i.test(label)) return parseMetric(label);
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
var currentUrl = window.location.href;
|
|
var path = String(window.location.pathname || '');
|
|
var loginRequired = path.indexOf('/login') >= 0
|
|
|| path.indexOf('/checkpoint/') >= 0
|
|
|| Boolean(document.querySelector('input[name="session_key"], form.login__form'));
|
|
var moreButtons = Array.from(document.querySelectorAll('button, a[role="button"]'))
|
|
.filter(function (el) {
|
|
return /see more|more/i.test(normalize(el.textContent))
|
|
|| /see more|more/i.test(normalize(el.getAttribute('aria-label')));
|
|
})
|
|
.slice(0, 8);
|
|
var cards = Array.from(document.querySelectorAll('article, .feed-shared-update-v2, .occludable-update, [role="listitem"]'));
|
|
var seen = new Set();
|
|
var posts = [];
|
|
var i;
|
|
var card;
|
|
var root;
|
|
var author;
|
|
var headline;
|
|
var text;
|
|
var postedAt;
|
|
var permalink;
|
|
var url;
|
|
var reactions;
|
|
var comments;
|
|
|
|
for (i = 0; i < moreButtons.length; i += 1) {
|
|
try { moreButtons[i].click(); } catch (err) {}
|
|
}
|
|
|
|
for (i = 0; i < cards.length; i += 1) {
|
|
card = cards[i];
|
|
root = card.closest('article, .feed-shared-update-v2, .occludable-update, [role="listitem"]') || card;
|
|
if (!root || seen.has(root)) continue;
|
|
seen.add(root);
|
|
|
|
if (String(root.getAttribute('role') || '') === 'listitem') {
|
|
var extracted = extractFromListItem(root);
|
|
if (extracted) posts.push(extracted);
|
|
continue;
|
|
}
|
|
|
|
author = normalize(
|
|
textOf(root, '.update-components-actor__title span[dir="ltr"]')
|
|
|| textOf(root, '.update-components-actor__title')
|
|
|| textOf(root, '[data-control-name="actor"] span[dir="ltr"]')
|
|
|| textOf(root, '[data-control-name="actor"]')
|
|
);
|
|
headline = normalize(
|
|
textOf(root, '.update-components-actor__description')
|
|
|| textOf(root, '.update-components-actor__sub-description')
|
|
);
|
|
text = normalize(
|
|
textOf(root, '.update-components-text span[dir="ltr"]')
|
|
|| textOf(root, '.update-components-text')
|
|
|| textOf(root, '.feed-shared-inline-show-more-text span[dir="ltr"]')
|
|
|| textOf(root, '.feed-shared-inline-show-more-text')
|
|
|| textOf(root, '[data-test-id="main-feed-activity-card"] .break-words')
|
|
);
|
|
postedAt = normalize(
|
|
textOf(root, '.update-components-actor__sub-description a')
|
|
|| textOf(root, '.update-components-actor__sub-description span[aria-hidden="true"]')
|
|
|| textOf(root, 'time')
|
|
);
|
|
permalink = root.querySelector('a[href*="/feed/update/"], a[href*="/posts/"], a[href*="/pulse/"]');
|
|
url = permalink ? permalink.href : '';
|
|
if (url && url.indexOf('/') === 0) url = new URL(url, currentUrl).toString();
|
|
reactions = parseMetric(
|
|
textOf(root, '.social-details-social-counts__reactions-count')
|
|
|| attrOf(root, '[aria-label*="reaction"]', 'aria-label')
|
|
|| attrOf(root, '[aria-label*="like"]', 'aria-label')
|
|
);
|
|
comments = commentMetric(root);
|
|
|
|
if (!author || !text) continue;
|
|
|
|
posts.push({
|
|
id: url || (author + '::' + postedAt + '::' + text.slice(0, 120)),
|
|
author: author,
|
|
author_url: hrefOf(root, 'a[href*="/in/"], a[href*="/company/"]'),
|
|
headline: headline,
|
|
text: text,
|
|
posted_at: postedAt,
|
|
reactions: reactions,
|
|
comments: comments,
|
|
url: url,
|
|
});
|
|
}
|
|
|
|
return { loginRequired: loginRequired, posts: posts };
|
|
})()`);
|
|
}
|
|
|
|
cli({
|
|
site: 'linkedin',
|
|
name: 'timeline',
|
|
description: 'Read LinkedIn home timeline posts',
|
|
domain: 'www.linkedin.com',
|
|
strategy: Strategy.COOKIE,
|
|
browser: true,
|
|
args: [
|
|
{ name: 'limit', type: 'int', default: 20, help: 'Number of posts to return (max 100)' },
|
|
],
|
|
columns: ['rank', 'author', 'author_url', 'headline', 'text', 'posted_at', 'reactions', 'comments', 'url'],
|
|
func: async (page, kwargs) => {
|
|
const limit = Math.max(1, Math.min(kwargs.limit ?? 20, 100));
|
|
|
|
await page.goto('https://www.linkedin.com/feed/');
|
|
await page.wait(4);
|
|
|
|
let posts: TimelinePost[] = [];
|
|
let sawLoginWall = false;
|
|
|
|
for (let i = 0; i < 6 && posts.length < limit; i++) {
|
|
const batch = await extractVisiblePosts(page);
|
|
if (batch?.loginRequired) sawLoginWall = true;
|
|
posts = mergeTimelinePosts(posts, Array.isArray(batch?.posts) ? batch.posts : []);
|
|
if (posts.length >= limit) break;
|
|
await page.autoScroll({ times: 1, delayMs: 1200 });
|
|
await page.wait(1);
|
|
}
|
|
|
|
if (sawLoginWall && posts.length === 0) {
|
|
throw new AuthRequiredError('linkedin.com', 'LinkedIn timeline requires an active signed-in browser session');
|
|
}
|
|
|
|
if (posts.length === 0) {
|
|
throw new EmptyResultError('linkedin timeline', 'Make sure your LinkedIn home feed is visible in the browser.');
|
|
}
|
|
|
|
return posts.slice(0, limit).map((post, index) => ({
|
|
rank: index + 1,
|
|
...post,
|
|
}));
|
|
},
|
|
});
|
|
|
|
export const __test__ = {
|
|
parseMetric,
|
|
buildPostId,
|
|
mergeTimelinePosts,
|
|
};
|