Jina AI skill for Claude: 10 URL-to-Markdown recipes

The prompt

Use the jina-ai skill. Read https://docs.python.org/3/library/asyncio-task.html with `python3 scripts/jina_tools.py read` and answer from the fetched content only: what is the difference between asyncio.create_task and asyncio.ensure_future, and which one does the page recommend for new code? Quote the exact sentence you based the answer on.

What slides.md looks like

$ python3 scripts/jina_tools.py read "https://docs.python.org/3/library/asyncio-task.html"
{
  "title": "Coroutines and Tasks — Python 3 documentation",
  "url": "https://docs.python.org/3/library/asyncio-task.html",
  "content_markdown": "## Creating Tasks\n\nasyncio.create_task(coro, *,
    name=None, context=None)\nWrap the coro coroutine into a Task and
    schedule its execution..."
}

Pairs with

The prompt

Use the jina-ai skill. Run `python3 scripts/jina_tools.py search "Node.js LTS schedule current active version"` and answer strictly from the results: which Node major versions are in Active LTS and Maintenance right now? List each claim with the source URL it came from. If the results disagree, say so instead of picking one.

What slides.md looks like

$ python3 scripts/jina_tools.py search "Node.js LTS schedule current active version"
[
  {
    "title": "Node.js — Releases",
    "url": "https://nodejs.org/en/about/previous-releases",
    "content_markdown": "## Release schedule\n\nMajor Node.js versions enter
      Current release status for six months..."
  },
  { "title": "...", "url": "...", "content_markdown": "..." }
]

Pairs with

The prompt

Use the jina-ai skill. I have urls.txt with one docs URL per line (about 50 lines). Write and run a bash loop that calls `python3 scripts/jina_tools.py read` on each URL, extracts content_markdown with jq, and saves it to corpus/<slugified-hostname-path>.md. Sleep 4 seconds between calls so we stay under the keyless rate limit. Print a final count of files written and flag any URL that returned empty content.

What slides.md looks like

while IFS= read -r url; do
  slug=$(printf '%s' "$url" | sed -E 's|https?://||; s|[^A-Za-z0-9]+|-|g')
  python3 scripts/jina_tools.py read "$url" \
    | jq -r '.content_markdown' > "corpus/$slug.md"
  sleep 4
done < urls.txt
ls corpus | wc -l   # 50

Pairs with

The prompt

Use the jina-ai skill. Read this article with `python3 scripts/jina_tools.py read <url>`, then produce a brief with exactly four sections: TL;DR (2 sentences), Core claims (bullet list, each with the supporting evidence the author gives), What's opinion vs. measurement (split them), and Action for our team (1-2 sentences, assume we run a Postgres-backed Rails monolith). Do not pad. If the article gives no measurements, say so under evidence.

What slides.md looks like

## TL;DR
The author migrated from sidekiq to pg-backed jobs and cut infra
spend; the win came from deleting Redis, not from job throughput.

## Core claims
- Queue latency stayed under 100ms at their scale (measured, p95 graph)
- Postgres SKIP LOCKED handles their 2k jobs/min (measured)

## Opinion vs. measurement
Opinion: "Redis is operational overhead you don't need."

Pairs with

The prompt

Use the jina-ai skill. Write snapshot.sh: it reads https://example-competitor.com/changelog and /pricing via `python3 scripts/jina_tools.py read`, writes each content_markdown to snapshots/changelog.md and snapshots/pricing.md, then runs git add + git commit -m "snapshot $(date +%F)". Then run it once and show me `git diff HEAD~1 -- snapshots/` if a previous snapshot exists. Summarize any diff in two sentences.

What slides.md looks like

$ git diff HEAD~1 -- snapshots/pricing.md
-| Team    | $49/mo  | 5 seats  |
+| Team    | $59/mo  | 5 seats  |
+| Scale   | $199/mo | 25 seats, SSO |

# Claude: "Team tier price increased and a new Scale tier
# with SSO appeared — their first enterprise motion."

Pairs with

The prompt

Use the jina-ai skill. Question: how do Postgres logical replication and pglogical differ for zero-downtime major-version upgrades? First run `python3 scripts/jina_tools.py search` on the question. Pick the 3 most authoritative result URLs (prefer postgresql.org and maintainer blogs), read each with the read command, and write a comparison: capabilities table, failure modes each source warns about, and a recommendation. Cite which source each row came from.

What slides.md looks like

python3 scripts/jina_tools.py search \
  "postgres logical replication vs pglogical major upgrade" > hits.json
jq -r '.[].url' hits.json | head -3 | while IFS= read -r u; do
  python3 scripts/jina_tools.py read "$u" >> sources.jsonl
done
# 3 full pages → one cited comparison table

Pairs with

The prompt

Use the jina-ai skill. Claim to verify: "SQLite is the most widely deployed database engine in the world." Search for it with `python3 scripts/jina_tools.py search`, read the two most authoritative results in full, and return a verdict block: VERDICT (supported / contradicted / unverifiable), the verbatim passage from each source that supports the verdict, the source URLs, and one sentence on how confident I should be. Do not soften the verdict.

What slides.md looks like

VERDICT: supported
Source 1: https://www.sqlite.org/mostdeployed.html
  "SQLite is likely used more than all other database
   engines combined."
Source 2: independent coverage repeating the claim with
  the same first-party origin.
Confidence: high for "most deployed", noting the primary
source is the SQLite project itself.

Pairs with

The prompt

Use the jina-ai skill. This page is a client-rendered SPA and curl returns an empty shell: https://app.example.com/public/changelog. Fetch it with `python3 scripts/jina_tools.py read` instead and confirm you got real content by printing the first three headings from content_markdown. Then summarize the last five releases listed.

What slides.md looks like

$ curl -s https://app.example.com/public/changelog | wc -c
1204    # <div id="root"></div> and a script tag

$ python3 scripts/jina_tools.py read "https://app.example.com/public/changelog" \
    | jq -r '.content_markdown' | grep '^#' | head -3
## v2.41 — Edge caching for media
## v2.40 — SSO domain enforcement
## v2.39 — Bulk export API

Pairs with

The prompt

Use the jina-ai skill. Here are the 12 links from this week's newsletter: [paste links]. Read each with `python3 scripts/jina_tools.py read`. Return an annotated bibliography sorted by relevance to building LLM agents in TypeScript: for each link, one line — title, 15-word summary, and a tag from {read-now, skim, skip}. End with the single link I should read first and why, in one sentence.

What slides.md looks like

1. [read-now] Streaming tool calls without buffering —
   pattern for incremental tool-call parsing in TS agents.
2. [skim] Postmortem: agent loop burned $400 — retry logic
   lesson, applies to any provider.
3. [skip] "Top 10 AI tools" — listicle, no code.
...
Read first: #1 — it solves the exact buffering bug in
your current agent runner.

Pairs with

The prompt

Use the jina-ai skill as reference. Write me a minimal fetch_markdown function for our TypeScript agent that GETs https://r.jina.ai/<target-url> with an Authorization: Bearer header from JINA_API_KEY and X-Return-Format: json, parses {data: {title, url, content}}, and throws a typed error on HTTP 451 (rate-limit/abuse) telling the caller to add an API key. Include the curl equivalent in a comment for debugging.

What slides.md looks like

# the whole integration, as curl:
curl "https://r.jina.ai/https://example.com/post" \
  -H "Authorization: Bearer $JINA_API_KEY" \
  -H "X-Return-Format: json"

{ "data": { "title": "...", "url": "...",
    "content": "## The post, as Markdown..." } }

Pairs with