Blocked Page Recovery

What it is

Blocked Page Recovery is a bundled Hermes Agent skill, version 1.0.0, MIT licensed, for the moment a fetch fails: 403 or 429, a Cloudflare "Just a moment..." page, a paywall, or a bot-detection interstitial. The instruction is short. Do not give up, and do not loop on the same URL.

Third-party services often hold a copy. Work down a ladder, cheapest first.

Who it targets

• Research runs that die on a WAF or a paywall and would otherwise be marked "no source."
• Anyone citing a page that had to be recovered. The skill's rule is that the copy carries a provenance, and you keep it.
• Not a substitute for the live page when the fact you need is a current price, a stock count, or breaking news.

What it does

The ladder on the skill page:

1. Wayback Machine, via archive.org's available API, which returns a snapshot URL and a timestamp.
2. archive.today, rotating domains: archive.ph, then .md, .li, .is.
3. Jina Reader, only if JINA_API_KEY is set. Anonymous access is described as dead.
4. An API-first pivot: look for /api/, /graphql, .json, or RSS on the same host after two or three blocked HTML attempts.
5. A real browser, last, because it is the expensive resort.

The bundled script runs that order and prints the first body it accepts, with provenance:

python3 scripts/recover_page.py "https://example.com/blocked-article" --json

How to cite what comes back

Route	Provenance	How the skill says to cite it
Wayback or archive.today	snapshot	Cite with the snapshot date. Never present it as the live page.
Jina Reader	live	Server-side re-render. Cite normally.
Live fetch or browser	live	Cite normally.

If you need current data, a snapshot is context. Say how old it is.

Dependencies

Hermes Agent. Path on the skill page: skills/web/blocked-page-recovery. Restore if the local tree is missing:

hermes skills reset blocked-page-recovery --restore

Python 3 for the script. JINA_API_KEY only if you want step 3. Related skill named on the page: grounded-citations.

Limitations

• Wayback fails for robots-blocked sites, URLs that were never crawled, and JS-only SPAs. A snapshot does not render the app.
• archive.today rate-limits hard. A 429 can still return several kilobytes of HTML. The skill says to validate the body, not the status code.
• Generic "web proxy" relays are called out as something not to use. Do not send cookies or Authorization headers through one. Provenance is unverifiable.
• Routes that lie with HTTP 200 are named: webcache.googleusercontent.com, AMP CDN hosts, and titles like "Just a moment," "Redirecting," or "Attention Required."
• CDX can return 503 under load. The skill says fall back to the available API instead of hammering it.
• This note is from the skill page fetched September 30, 2026. The script was not run for this entry.