Skip to content

Latest commit

 

History

1 Commit

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 

Repository files navigation

Web intelligence

Five ways to get something off the web, cheapest first.

Most tools reach for the most expensive retrieval method every time. This picks the cheapest route that will actually answer the question, and steps up only when it has to.

Tier Route Time Cost Used when
1 Training data instant free The answer is not time-sensitive
2 WebFetch seconds free A specific URL on a cooperative site
3 Firecrawl API seconds 1 credit JS-heavy sites, or WebFetch failed
4 Codex + small model 1-3 min free Targeted current-info questions
5 Codex + search 20-30 min free Broad investigation, confirm first

What you need

  • Tiers 1 and 2 need nothing at all. You can use this immediately.
  • FIRECRAWL_API_KEY for tier 3.
  • The Codex CLI, signed in, for tiers 4 and 5.

Known traps

These are real things that cost time, written down so they cost you none.

  • Tier 5 can take half an hour. The skill is written to confirm with you before starting one, rather than disappearing for thirty minutes.

Portability

Fully portable, and it degrades tier by tier on its own. If you only want to read one skill here to see how graceful fallback should work, read this one.

Installing it

This is a folder of markdown instructions plus its helper scripts. Any AI that can read instructions from a file can use it. Pick your tool:

Claude Code

cp -r web-intelligence ~/.claude/skills/

Or drop it in <your-project>/.claude/skills/ to keep it project-local.

Codex, Antigravity, Cursor, or anything else that reads local files

Put the folder anywhere in your project and point your agent at it. In Codex, adding a line to AGENTS.md is enough:

The web-intelligence skill lives in ./web-intelligence. Read web-intelligence/SKILL.md before using it.

Claude.ai or another browser chat

Paste the contents of SKILL.md into the chat. The tiers that need no keys will work; the ones that shell out will not.

Anything else

SKILL.md is plain markdown with YAML frontmatter (name, description). If your platform rejects unknown frontmatter keys, delete the model: line and it will load.

Before you run it

Do not blindly trust skills you download from the internet, including mine. This one ships scripts that run on your machine. Check it first, which your AI can do in about a minute:

Read every file in this folder, including the scripts. Tell me plainly: does anything
here send my data anywhere, contact the internet, run something I would not expect, or
ask for credentials? Quote the exact lines that made you say so. If it is clean, say so.

If the answer worries you, do not run it.

Where this came from

This is a real working skill, pulled out of a private knowledge vault and anonymised for release. It is not a demo. The Session History section inside SKILL.md keeps the bugs and dead ends found while building it, because those are usually the most useful part.

Licence

MIT. Do what you like with it.

Built by JP Pieterse.

About

Five ways to get something off the web, cheapest first. Degrades tier by tier.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages