Blog on Restish

Scripting REST APIs Without Fragile curl Loops

Tue, 30 Jun 2026 00:00:00 +0000

Somewhere in your infrastructure there is a shell script with a loop like this:

url="https://api.example.com/items?page=1"
while [ -n "$url" ]; do
 resp=$(curl -s "$url")
 echo "$resp" | jq -r '.items[].id'
 url=$(echo "$resp" | jq -r '.links.next // empty')
done

It works, mostly. It has also quietly accumulated a list of failure modes that nobody will notice until a bad day:

A 500 response exits with status 0, so the error page flows into jq, which prints nothing, and the script “succeeds” with empty output.
There is no retry, so one transient network blip fails the whole job — or worse, half of it.
There is no timeout, so a hung connection hangs the cron job behind it.
There is no page bound, so a pagination bug upstream turns the loop into an accidental load test.
If next ever points somewhere unexpected — another host, an attacker- influenced URL in a response body — the loop follows it without a thought.

None of these are exotic. They are the standard tax on hand-rolling HTTP plumbing in shell, paid one incident at a time. This post is about paying it once, in the tool, instead.

TOON: Token-Efficient API Output for LLM Workflows

Mon, 15 Jun 2026 00:00:00 +0000

When a person reads an API response, formatting is free. When an LLM agent reads one, every character is metered. The response lands in a context window, the context window is billed by the token, and the format you chose decides how many tokens the same data costs.

JSON spends a lot of tokens on that job. A list of one hundred records repeats every key one hundred times, wraps every string in quotes, and spends tokens on braces and brackets that a model does not need to understand a table of data.

Turn an OpenAPI Spec Into a CLI Without Generating Code

Mon, 08 Jun 2026 00:00:00 +0000

OpenAPI is usually treated as an input to something else. Generate an SDK. Render Swagger UI. Publish reference docs. Feed a contract test. All of those are good jobs for a spec.

But there is another useful job hiding in the same document: teach the terminal how to work with the API.

That does not have to mean generating a new repository, choosing a language runtime, publishing a package, and keeping a client binary in sync with every operation change. A CLI can load an OpenAPI description at runtime, cache what it needs, and turn repeated API work into commands without making users rebuild a client.

Restish v2: A CLI for REST APIs, Rebuilt

Sat, 30 May 2026 00:00:00 +0000

Most APIs already know more about themselves than the command line does. They have schemas, auth rules, pagination links, content types, examples, and often a full OpenAPI description. Then we copy a curl command into a terminal and teach all of that context back to the shell one flag at a time.

curl is still the universal primitive. Restish is for the moment after that: when a URL becomes a workflow, and the API should start feeling like a small native CLI instead of a string you keep rebuilding.