agentsclimarketplace

Jina ai

Skill archibate/agent-skills/skills/jina-ai

Web search with time-window, region/language, and `site:` filters, academic papers (arXiv/SSRN), PDF table/figure extraction, BibTeX, image search, web page reading, embeddings, reranking, classification, and deduplication (text or images) via Jina AI. Use when searching web content, finding academic papers, or extracting figures from PDFs. Prefer this over WebSearch for better results.From its SKILL.md

Install
npx -y skills add archibate/agent-skills --skill jina-ai

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 16 stars16 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

9.0 KB, ~2.5k tokens by cl100k_base, as published. Nobody here has run it

Jina AI

Use the jina CLI for all Jina AI APIs — search, read, embed, rerank, classify, dedup, screenshot, BibTeX, PDF figures. Composable with Unix pipes.

Setup

One-time install (skip if command -v jina returns a path):

uv tool install jina-cli --with 'httpx[socks]'

JINA_API_KEY must be set in the environment. Get one at https://jina.ai/?sui=apikey.

Commands

CommandFunction
jina read URLExtract clean markdown from a web page
jina search QUERYWeb search (also --arxiv, --ssrn, --images, --blog)
jina embed TEXTGenerate embeddings
jina rerank QUERYRerank stdin documents by relevance
jina classify TEXT --labels a,b,cClassify text into labels
jina dedup [-k N]Deduplicate stdin text lines
jina screenshot URLCapture a screenshot of a URL
jina bibtex QUERYSearch BibTeX citations (DBLP + Semantic Scholar)
jina expand QUERYExpand a query into related queries
jina pdf URL|ARXIV_IDExtract figures/tables/equations from a PDF
jina datetime URLGuess publish/update date of a URL
jina primerSession context (time, location, network)
scripts/dedup_images.py PATH_OR_URL ...Deduplicate images by visual similarity (CLIP v2) — see Image dedup

Most jina subcommands support --json for structured output and --api-key to override $JINA_API_KEY.

Pipes

Commands read stdin and write stdout, so chain them:

# Search and rerank
jina search "transformer models" | jina rerank "efficient inference"

# Read multiple URLs (one per line on stdin)
cat urls.txt | jina read

# Search, then deduplicate near-identical results
jina search "attention mechanism" | jina dedup

# Expand a query, then search the first variant
jina expand "climate change" | head -1 | xargs -I {} jina search "{}"

# Get JSON, slice with jq
jina search --arxiv "BERT" --json | jq -r '.results[].title'

For batch fan-out where the subcommand only takes one input (e.g. search, bibtex), launch parallel Bash calls or use xargs -P:

printf '%s\n' "query A" "query B" "query C" | xargs -P 3 -I {} jina search "{}" --json

To fan a single query into 5 diverse parallel searches:

jina expand "LLM" | xargs -P 5 -I {} jina search "{}"

Usage

Read web pages

jina read https://example.com
jina read https://example.com --links --images

If jina read not working: Fallback to use $read-url skill instead.

Search

jina search "what is BERT"
jina search --arxiv "attention mechanism" -n 10
jina search --ssrn "corporate governance"
jina search --images "neural network diagram"
jina search --blog "embeddings"
jina search "AI news" --time d                  # past day (h|d|w|m|y)
jina search "深度学习" --gl cn --hl zh-cn       # Chinese region/language
jina search "LLM" --location "Shanghai"
jina search "site:github.com claude code hooks" # Google-style operators pass through (site:, "quoted", -exclude, OR, intitle:, inurl:)

Embed

jina embed "hello world"
jina embed "text1" "text2" "text3"
cat texts.txt | jina embed --json
jina embed "hello" --model jina-embeddings-v5-text-small --task retrieval.query

Rerank

cat docs.txt | jina rerank "machine learning"
jina search "AI" | jina rerank "embeddings" --top-n 5

Classify

jina classify "I love this product" --labels positive,negative,neutral
echo "stock prices rose sharply" | jina classify --labels business,sports,tech
cat texts.txt | jina classify --labels cat1,cat2,cat3 --json

Deduplicate (text)

cat items.txt | jina dedup
cat items.txt | jina dedup -k 10

Image dedup

jina dedup is text-only. For visual deduplication of images, use the bundled script:

scripts/dedup_images.py *.png                          # local paths, default keeps n//2
scripts/dedup_images.py -k 5 --json img1.jpg img2.jpg
ls images/*.png | scripts/dedup_images.py -k 3
scripts/dedup_images.py https://example.com/a.png /tmp/b.png   # mix URLs + paths

It calls https://api.jina.ai/v1/embeddings with model jina-clip-v2 and runs greedy farthest-point sampling on cosine similarity. Local paths are read and base64-encoded; http(s)://… and data: URIs pass through. Prefer local paths — Jina's URL fetcher cannot reach some hot-link-protected hosts (e.g. Wikimedia, certain CDNs).

Screenshot

jina screenshot https://example.com                    # prints screenshot URL
jina screenshot https://example.com -o page.png        # saves to file
jina screenshot https://example.com --full-page -o page.jpg

BibTeX

jina bibtex "attention is all you need"
jina bibtex "transformer" --author Vaswani --year 2017

PDF figure extraction

jina pdf https://arxiv.org/pdf/2301.12345
jina pdf 2301.12345                                    # arXiv ID shorthand
jina pdf https://example.com/paper.pdf --type figure,table

Read academic papers (arXiv)

The URL form determines what you get — pick deliberately:

GoalCommand
Abstract + metadata onlyjina read arxiv.org/abs/<ID>
Full paper body as markdownjina read arxiv.org/pdf/<ID>
Figures / tables / equationsjina pdf <ID>
BibTeX citationjina bibtex "<title>"
Save raw PDFcurl -L -o paper.pdf arxiv.org/pdf/<ID>
# Find candidates
jina search --arxiv "diffusion transformer" -n 10 --json | jq -r '.results[] | "\(.title)\t\(.url)"'

# Abstract + metadata (~2 KB)
jina read https://arxiv.org/abs/1706.03762

# Full paper as markdown (~40 KB for a conference paper)
jina read https://arxiv.org/pdf/1706.03762 > paper.md

# Figures and tables
jina pdf 1706.03762 --type figure,table

# Citation
jina bibtex "Attention Is All You Need" --author Vaswani

Read academic papers (SSRN)

SSRN sits behind a Cloudflare bot challenge — jina read and plain curl return 403 on both abstract pages (papers.cfm?abstract_id=…) and PDF endpoints (Delivery.cfm).

What works (tiered by effort):

  1. jina search --ssrn snippets. Each hit's JSON record has title, abstract excerpt, date, and ssrn_id. Often sufficient for triage and citation scaffolding.
  2. scrapling skill on the abstract page. scrapling extract stealthy-fetch --solve-cloudflare returns the full abstract, authors, citation block, and the resolved PDF download URL as markdown.
  3. PDF body text. scrapling's CLI stealthy-fetch does not succeed on the Delivery.cfm URL — its Cloudflare DOM solver expects an HTML response, not a binary PDF. Getting the PDF text needs a Python StealthySession that hits the abstract page first, reuses cookies to download the PDF, then feeds the bytes to the pdf skill.
  4. jina bibtex "<title>" resolves citations independently of SSRN.
# Tier 1: search snippets
jina search --ssrn "corporate governance" -n 5 --json | jq -r '.results[] | "\(.ssrn_id)\t\(.title)\n  \(.snippet)"'

# Tier 2: abstract page via scrapling (Cloudflare bypass)
scrapling extract stealthy-fetch \
  "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=<SSRN_ID>" \
  /tmp/ssrn.md --solve-cloudflare --timeout 60000

Other

jina datetime https://example.com/article              # guess publish date
jina expand "machine learning optimization"            # query variants
jina primer                                            # session context

JSON output and exit codes

All data-returning subcommands support --json for structured output (pipe to jq).

CodeMeaning
0Success
1User/input error (missing args, bad input, missing API key)
2API/server error (network, timeout, HTTP error)
130Interrupted (Ctrl+C)
jina search "query" && echo "ok" || echo "failed: $?"

Environment

VariableDescription
JINA_API_KEYRequired for most commands

Tool Selection Guide

ScenarioTool
Read a web page$read-url skill
Find STEM papersjina search --arxiv
Find social-science / finance papersjina search --ssrn
Read paper abstract / metadatajina read (/abs/ URL)
Download full paper as markdownjina read (/pdf/ URL)
Save raw PDFcurl -L
Generic web searchjina search
Fallback if jina service unreachableWebSearch (built-in)

Tips

  • For Chinese results, set --gl cn --hl zh-cn; for date-bounded results, --time w (past week).
  • Use --json when parsing output; default text is for humans and Unix pipes.
  • Errors go to stderr with a fix hint; check $? (or use &&/||) rather than parsing stderr.

What ships with it: 1 file

4.8 KB alongside SKILL.md, 1 of them executable

scripts/

Keep looking

Skills are one crate of 608,119. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.