Find usable copies of web pages when normal fetching fails. The skill tries archives and alternative retrieval routes, checks that results contain actual page content, and preserves source details and snapshot dates for citation.
中文简介正在准备中。当前显示英语简介。
Use when a fetch fails: 403/429, paywall, WAF, bot wall.
安装前可以查看交给 AI 代理的指令内容。
When a page won't fetch — 403/429, Cloudflare "Just a moment...", a paywall, or a bot-detection interstitial — don't give up and don't loop on the same URL. Third-party services often hold a copy of the page. Work down this ladder, cheapest first.
1. Wayback Machine — archive.org "available" API (snapshot + timestamp)
2. archive.today — domain rotation: archive.ph → .md → .li → .is
3. Jina Reader — only if JINA_API_KEY is set (live server-side render)
4. API-first pivot — look for /api/, /graphql, .json, or RSS on the same host
5. Real browser — browser tool as the last, most expensive resort
Run it in one shot with the bundled script:
python3 scripts/recover_page.py "https://example.com/blocked-article" --json
The script tries each route in order, validates every body (see "Fake successes" below), and prints the first genuine hit with its provenance.
Every recovered copy carries a provenance you MUST preserve when citing:
| Route | Provenance | How to cite |
|---|---|---|
| Wayback / archive.today | snapshot | Cite WITH the snapshot date: "as archived 2026-08-06". Never present a snapshot as the live page — it may be stale. |
| Jina Reader | live | Server-side re-render of the live page; cite normally. |
| Live fetch / browser | live | Cite normally. |
If the user needs current data (prices, availability, breaking news), a snapshot is context, not an answer — say so explicitly and note its age.
# Discovery: returns closest snapshot URL + timestamp as JSON
curl -sL "https://archive.org/wayback/available?url={URL}"
# Then fetch archived_snapshots.closest.url
For enumerating many snapshots (or recovering deleted pages), the CDX index:
curl -sL "https://web.archive.org/cdx/search/cdx?url={URL}&output=json&limit=10"
CDX intermittently returns 503 under load — if it does, fall back to the
available API; don't retry-hammer it.
Works for: any publicly crawled URL. Fails for: robots-blocked sites, never-crawled URLs, JS-only SPAs (snapshots don't render).
User-submitted archives — often has paywalled news articles Wayback lacks. Rate-limits aggressively (429) and rotates domains, so iterate:
for d in archive.ph archive.md archive.li archive.is; do
curl -sL --max-time 20 "https://$d/newest/{URL}" -o ~/.hermes/cache/scratch/page.html \
-w "%{http_code}" && break
done
Validate the body, not the status code — a 429 still ships several KB of rate-limit HTML that looks like a success to a size check alone.
r.jina.ai re-renders the live page in a real browser server-side and
returns markdown. Anonymous access is dead (401 → Turnstile); a key is
required:
curl -s -H "Authorization: Bearer $JINA_API_KEY" "https://r.jina.ai/{URL}"
Handles JS SPAs that archives can't. Skip this route entirely when the env var is unset.
WAFs protect the HTML surface far more aggressively than the data endpoints behind it. After 2-3 blocked attempts on a site, stop fighting the HTML and look for:
/api/..., /graphql, or .json variants of the page URL/feed, /rss, <link rel="alternate"> in any copy
you did recover)/sitemap.xml) revealing canonical URLs that may not be gatedThese return HTTP 200 with a plausible body that is NOT the page. The script rejects them automatically; reject them manually too:
webcache.googleusercontent.com
returns 200 + tens of KB, but it's a Google Search interstitial with a JS
redirect, not a cache. Never use it.*.cdn.ampproject.org) mostly return a ~300-byte
<title>Redirecting</title> meta-refresh stub pointing back at the
original (blocked) URL. Treating that as success creates a fetch loop.Detection heuristics the script applies: body under a per-route byte floor; meta-refresh/JS-redirect stubs whose target is the original host; interstitial titles ("Just a moment", "Redirecting", "Google Search", "Attention Required").
Generic "web proxy" relays are man-in-the-middle by construction. Never send cookies or Authorization headers through one, and don't use them for anything the user will rely on — provenance is unverifiable. Prefer archives, which at least timestamp their copies.
还没有评价。欢迎分享你的使用感受。
无需登录即可发表。欢迎分享使用感受、提问，或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
Set up the 1Password CLI and use stored passwords or API keys in commands and configuration templates. Choose an authentication method, retrieve secret references, and pass credentials to processes through environment variables.
中文简介正在准备中。当前显示英语简介。
Build an Excel workbook linking the income statement, balance sheet, and cash flow statement. Populate historical figures, create formula-based forecasts, compare scenarios, and check that balances and cash totals agree across all three statements.
中文简介正在准备中。当前显示英语简介。
Adapt existing PyTorch training scripts to use multiple GPUs or machines with limited code changes. The skill guides configuration, launch commands, mixed precision, which uses lower-precision numbers, and DeepSpeed or FSDP settings for distributing training work and memory.
中文简介正在准备中。当前显示英语简介。
Configure Hermes to run AI models through Actual Computer on your own hardware. Set up either the hosted relay or a local connection, select models, verify responses, and troubleshoot request errors or empty output.
中文简介正在准备中。当前显示英语简介。
Test an app from the perspective of a frustrated, less technical user. Find confusing steps and barriers, separate actionable usability problems from personal complaints, and produce screenshots, suggested fixes, and prioritized tickets.
中文简介正在准备中。当前显示英语简介。
Resolve Git merge conflicts between two agents by comparing both changes and their stated goals. Combine compatible work, explain choices where designs differ, and verify the merged result with project builds or tests.
中文简介正在准备中。当前显示英语简介。