[h3qing]

New linkedin-session perf-data adapter for /cheat-retro, mirroring the douyin-session architecture (Playwright persistent login + read rendered DOM — no signature reversing, no public API).

What it scrapes

LinkedIn's single-post analytics page (/analytics/post-summary/urn:li:activity:<id>/) is author-only and SSR/inline — there's no public API and no stable voyager XHR to intercept, so pure HTTP / requests (and WebFetch) can't reach it. The adapter logs in once (li_at cookie persisted to .auth-linkedin/), navigates to the post-summary page, and parses the rendered DOM text by known label anchors.

Captures 10 metrics for one post: impressions, members reached, reactions, comments, reposts, saves, sends, social engagements, profile-views-from-post, followers-from-post, plus the post body.

Matches the perf-data convention

• bash run.sh <activity_id_or_url> <video_folder> [<script_path>] → writes <video_folder>/report.md, with the same exit codes as douyin-session (0 success · 1 login required · 2 missing playwright · 3 other) so /cheat-retro degrades gracefully to manual on any failure.
• paths.py honors CHEAT_PROJECT_ROOT / CHEAT_VIDEOS_DIR like the other adapters; auth lives in its own .auth-linkedin/ to avoid cookie cross-talk.
• report.md matches the existing adapters' shape: meta (author / age / link) → data snapshot (the 10 metrics + derived ratios: 反应率 / 评论率 / 转发率 / 社交互动率) → post body → original script → comments note.
• review.py CLI matches the login / video <id> [script] contract.
• Accepts a full post URL (…/feed/update/urn:li:activity:<id>/) or a bare activity_id and auto-extracts the id.
• Adds .gitattributes (*.sh / *.py eol=lf) like bilibili-stat, to keep Windows-CRLF from breaking the scripts.

Bilingual JP/EN parser (honest maintenance note)

LinkedIn randomly flips the UI language between Japanese and English within a single session (a label shows up as Impressions one load and インプレッション数 the next). So extract.py's POST_METRICS stores both label aliases per metric and tries both. If you hit a third language, add the label to POST_METRICS — the aliases are a list. test_extract.py exercises both JP and EN with synthetic fixtures (no real metrics committed). The parser is DOM-text + language dependent, so it's marked stability ★★ and needs occasional upkeep; the debug dump (.cheat-cache/linkedin-session-debug/post_<id>.txt) is the first place to look when a field comes back None.

Install / usage

cd ~/my-channel
python3 -m venv .venv && source .venv/bin/activate
pip install -r adapters/perf-data/linkedin-session/requirements.txt   # playwright>=1.44
playwright install chromium
python adapters/perf-data/linkedin-session/crawler.py login           # log in once → .auth-linkedin/
# then /cheat-retro calls run.sh automatically; manual test:
python adapters/perf-data/linkedin-session/review.py video 7470493738918920193

How to get the activity_id: it's the number in a post URL https://www.linkedin.com/feed/update/urn:li:activity:<id>/.

Wiring

• skills/cheat-retro/SKILL.md: adds the linkedin row to the adapter routing table + a linkedin-session 的特殊处理 block (same pattern as the douyin / xhs / bilibili PRs).
• skills/cheat-init/SKILL.md: adds LinkedIn to the Q2.1 platform menu + the enabled_perf_adapters mapping.
• .gitignore + templates/gitignore.template: add .auth-linkedin/ so the login cookie is never committed.

Tested

End-to-end on a real post (the example activity_id above): run.sh produces a sane report.md in the requested folder (auto-named-folder move works), and the unit tests pass for both JP and EN label sets. No auth, debug output, or __pycache__ is committed.

🤖 Generated with Claude Code

[Jooonnn]

感谢这个 PR，整体质量很高 —— 架构忠实复刻了 douyin-session，JP/EN 双语 label parser 设计得当，cookie/auth 保密处理干净（.auth-linkedin/ 已进 .gitignore 和模板，无 secret/真实指标/__pycache__ 入库）。已合并 ✅

一个建议后续优化（非阻塞）：

• run.sh 的 PROJECT_ROOT 定位与其它适配器不一致：本 PR 用 dirname(dirname(realpath VIDEO_FOLDER))，硬假设视频目录正好在 <root>/videos/<folder> 两层之下。fix: 全仓代码审查 — 修复 Python / Shell / 协议 / 配置共 34 个 bug #49 之后 douyin/xhs/bilibili 都改成了向上查找 .cheat-state.json（不依赖嵌套深度 + 校验真项目根 + 找不到 exit 3）。当前写法若层级≠2 会算出错误 root，再因找不到 .auth-linkedin/ 报一个误导性的 exit 1 "Not logged in"。建议对齐到 .cheat-state.json 向上查找，保持优雅降级体验。