himself65/opencli-reader

opencli Reader (Generic Fallback, Read-Only)

Generic fallback for any source opencli supports via its adapter registry (90+ sites, growing). Use this skill only when no dedicated finance-skill covers the source — the specialized skills (twitter-reader, linkedin-reader, discord-reader, telegram-reader, yc-reader) are always preferred when the request matches one of them.

This skill is read-only. Write commands that opencli exposes (post, like, comment, send, save, upvote, subscribe, follow, delete, reply-dm, etc.) must not be invoked.

---

Step 1: Decide Whether to Use This Skill

Only use this skill if the request cannot be handled by a more specific skill.

| If the user asks about… | Use this skill instead | |---|---| | Twitter/X | twitter-reader | | LinkedIn | linkedin-reader | | Discord | discord-reader | | Telegram | telegram-reader | | Y Combinator | yc-reader | | Anything else opencli supports (Yahoo Finance, Bloomberg, Reuters, Reddit, HackerNews, Eastmoney, Xueqiu, Substack, arXiv, etc.) | this skill |

If the source is not in opencli's registry either, stop and tell the user the request isn't covered — don't fall back to ad-hoc scraping.

---

Step 2: Ensure opencli Is Ready

Current environment status:

!`(command -v opencli && opencli doctor 2>&1 | head -5 && echo "READY" || echo "SETUP_NEEDED") 2>/dev/null || echo "NOT_INSTALLED"`

If NOT_INSTALLED:

npm install -g @jackwener/opencli

If SETUP_NEEDED, guide the user through Browser Bridge setup (only required for adapters whose strategy is COOKIE, HEADER, INTERCEPT, or UI — PUBLIC and LOCAL adapters work without a browser):

1. Download the latest opencli-extension-v{version}.zip from the GitHub Releases page
2. Unzip it, open chrome://extensions in Chrome, enable Developer mode
3. Click Load unpacked and select the unzipped folder
4. Make sure Chrome is logged into the target site, then re-run opencli doctor

Requires Node.js >= 21 (or Bun >= 1.0).

---

Step 3: Discover the Right Command

Do not guess command names or flags — the registry has 500+ commands and changes weekly. Instead:

# Full registry (grouped by site), machine-readable JSON
opencli list -f json

# Filter to a site
opencli list | grep -i <site>

# Site-level help (all commands + flags)
opencli <site> --help

# Command-level help (positional args + flags + defaults)
opencli <site> <command> --help

The opencli list -f json entry for each command includes:

• site — adapter namespace (e.g., yahoo-finance)
• name — subcommand (e.g., quote)
• strategy — PUBLIC / COOKIE / HEADER / INTERCEPT / UI / LOCAL — tells you if a browser login is needed
• description, args, columns — canonical metadata

Use opencli list -f json as the source of truth. Never paste a site list into the plan from memory; adapters are added every week.

Quick map of the most common finance / research sources

The table below is a shortlist, not exhaustive — always confirm with opencli <site> --help.

| Source | Site slug | Common commands | |---|---|---| | Yahoo Finance | yahoo-finance | quote | | Bloomberg | bloomberg | markets, economics, industries, tech, politics, opinions, news, businessweek, feeds, main | | Reuters | reuters | search | | Eastmoney (东方财富) | eastmoney | quote, rank, kline, sectors, etf, holders, money-flow, northbound, longhu, kuaixun, convertible, index-board, announcement, hot-rank | | Xueqiu (雪球) | xueqiu | stock, hot-stock, hot, feed, comments, watchlist, search, groups, fund-snapshot, fund-holdings, earnings-date, kline | | Sinafinance | sinafinance | (see --help) | | TDX / THS | tdx, ths | (see --help) | | Barchart (options) | barchart | quote, options, flow, greeks | | Reddit | reddit | hot, popular, frontpage, search, subreddit, read, user, user-posts, user-comments, saved | | HackerNews | hackernews | top, best, new, ask, show, jobs, user, search | | Substack | substack | feed, publication, search | | Medium | medium | (see --help) | | arXiv | arxiv | (see --help) | | Google Scholar | google-scholar | (see --help) | | Weibo | weibo | (see --help) | | Bilibili | bilibili | hot, video + more | | Xiaohongshu (小红书) | xiaohongshu | (see --help) | | Rednote (小红书 international) | rednote | (see --help — mirrors xiaohongshu) | | Zhihu | zhihu | (see --help) | | Tieba (百度贴吧) | tieba | (see --help) | | Hupu (虎扑) | hupu | (see --help) | | Xianyu (闲鱼) | xianyu | (see --help) | | 1688 | 1688 | (see --help) | | Gitee | gitee | (see --help) | | Quark | quark | (see --help) | | Baidu Scholar | baidu-scholar | (see --help) | | Nowcoder | nowcoder | (see --help) | | Wanfang | wanfang | (see --help) | | Doubao (豆包) | doubao | (see --help) | | Yuanbao (腾讯元宝) | yuanbao | (see --help) | | Google Gemini | gemini | (see --help) | | NotebookLM | notebooklm | (see --help) | | Claude | claude | (see --help) | | 36kr | 36kr | (see --help) | | Jike | jike | (see --help) | | Bluesky | bluesky | (see --help) | | Apple Podcasts | apple-podcasts | (see --help) | | Xiaoyuzhou (podcasts) | xiaoyuzhou | (see --help) | | Spotify | spotify | (see --help) | | YouTube | youtube | (see --help) | | Weixin Official Account | weixin | (see --help — drafts is read; create-draft is write) | | Toutiao | toutiao | articles | | Government policy / law | gov-policy, gov-law | (see --help) | | Web download / reader | web | read, download |

For anything not listed, run opencli list -f json and filter.

---

Step 4: Check the Adapter's Strategy Before Running

Run opencli list -f json (or opencli <site> <command> --help) and read the strategy field:

| Strategy | What it means | Preconditions | |---|---|---| | PUBLIC | Pure HTTP; no browser needed | None | | LOCAL | Talks to a local endpoint | Local service running | | COOKIE / HEADER | Reuses your Chrome login for the site | Chrome logged into the site + Browser Bridge extension loaded | | INTERCEPT | Opens an automation window to capture a signed request | Same as COOKIE; be patient — may take several seconds | | UI | Full DOM interaction | Same as COOKIE; slowest; results depend on the site's current layout |

If the user doesn't have a login and the adapter's strategy is not PUBLIC / LOCAL, tell them they need to log into the site in Chrome before retrying.

---

Step 5: Execute the Command

General pattern

opencli <site> <command> [positional-args] [flags] -f json

Universal flags

| Flag | Effect | |---|---| | -f json | Structured JSON — always prefer this for agent processing | | -f yaml / -f csv / -f md / -f table / -f plain | Other formats | | -v | Verbose logging (also sets OPENCLI_VERBOSE=1) | | --live | Keep the automation window open after the command (browser-backed adapters only) | | --focus | Open the automation window in the foreground (browser-backed adapters only) |

Command-specific flags (--limit, --filter, --type, etc.) are not universal — always check opencli <site> <command> --help.

Examples

# Yahoo Finance quote (PUBLIC)
opencli yahoo-finance quote AAPL -f json

# Reddit hot posts in a subreddit (COOKIE or PUBLIC depending on subreddit)
opencli reddit subreddit wallstreetbets --limit 20 -f json
opencli reddit search "SPY options" --limit 15 -f json

# HackerNews top (PUBLIC)
opencli hackernews top --limit 20 -f json

# Eastmoney hot rank (PUBLIC)
opencli eastmoney hot-rank -f json

# Xueqiu hot stocks (PUBLIC or COOKIE)
opencli xueqiu hot-stock -f json
opencli xueqiu stock SH600519 -f json

# Bloomberg markets headlines (COOKIE)
opencli bloomberg markets -f json

# arXiv paper search (PUBLIC)
opencli arxiv search "volatility surface" --limit 10 -f json

# Substack feed
opencli substack feed --limit 20 -f json

# Web page → readable markdown (PUBLIC)
opencli web read "https://example.com/article" -f json

Key rules

1. Always use opencli <site> <command> --help before constructing a command you haven't run this session — don't assume flag names.
2. Use -f json for programmatic processing.
3. Start with a small --limit (10–20) to validate the shape before pulling more.
4. Check strategy before running a browser-backed adapter — if the user isn't logged in, a COOKIE / UI adapter will fail.
5. NEVER execute write operations. Common write command names to avoid across adapters: post, reply, comment, like, unlike, upvote, save, subscribe, unsubscribe, follow, unfollow, block, unblock, delete, bookmark, unbookmark, send, create-draft, reply-dm, accept. If you're unsure whether a command is read or write, check the description in opencli list -f json; if it suggests a mutation, skip it.

---

Step 6: Handle Failures

If a command returns empty or errors out, the site may have changed its selectors / API. opencli has a built-in self-repair loop:

# Re-run with diagnostic context
OPENCLI_DIAGNOSTIC=1 opencli <site> <command> <args>

This emits a structured RepairContext that identifies the failing adapter's source path. Possible responses:

1. If the user has the opencli-autofix skill installed, tell them to run that skill.
2. If not, suggest they file an issue at https://github.com/jackwener/opencli/issues with the RepairContext output.
3. Don't silently fall back to hand-rolled scraping — that hides the bug from the upstream registry.

Rate limits on the target site can also cause empty results; wait and retry.

---

Step 7: Present the Results

1. Summarize the data for the user's actual question, don't just dump the raw JSON.
2. Include source attribution — site name + URL for each item where available.
3. For market data, surface price / % change / volume / market cap and flag anomalies.
4. For news/posts, highlight headlines, timestamps, and key quotes.
5. For research (arXiv, Scholar), include title, authors, abstract, and link.
6. Treat browser sessions as private — never echo CDP endpoints, cookies, or auth tokens.

---

Reference Files

• references/discovery.md — How to navigate opencli list, opencli <site> --help, and the JSON schema of registry entries
• references/finance-sources.md — Detailed notes on the finance-heavy adapters (Yahoo Finance, Bloomberg, Eastmoney, Xueqiu, Barchart, Reuters, Reddit, HackerNews) and which commands are read vs write

Read these reference files when you need concrete examples for a specific site, or when the user asks for a capability not covered by one of the dedicated readers.