Question 1

What is this hook for?

Accepted Answer

It fires when you write code containing an LLM call or a hand-written classifier and injects a three-way test asking whether the decision needs a model at all. It never blocks the edit, and the README says it recommends plain code at least as often as it recommends a model.

Question 2

What are the three bands of the test?

Accepted Answer

One: a regex, DNS lookup or database query can answer it, so write code instead of calling a model. Two: it needs multi-step reasoning, domain knowledge or generated prose, so use a frontier model. Three: a sensible person answers it in under a second from text you can show them — the System One band where Jev lives.

Question 3

How do I install the hook?

Accepted Answer

Copy hooks/typesafe-check.sh into .claude/hooks with curl or a manual copy, chmod +x it, then merge settings.example.json into .claude/settings.json. If a PreToolUse block already exists, add the hook to its hooks array rather than replacing it.

Question 4

How do I verify it is wired up?

Accepted Answer

The README suggests a jq query against .claude/settings.json for the Write|Edit matcher's command; exit 0 with a printed command means it is registered. If it does not fire, open /hooks once or restart the session, since Claude Code only watches directories that had a settings file at session start.

Question 5

Which files and function names trigger it?

Accepted Answer

Only .ts .tsx .cts .mts .js .jsx .cjs .mjs and .py files. Trigger classes match camelCase decision-shaped names such as parseIntent, getSentiment and assessSentiment; measured on 22 names, 8/8 camelCase decision names fired while SCREAMING_CASE constants and ordinary functions did not.

Question 6

Does the hook itself call Jev or send my code anywhere?

Accepted Answer

The captured README shows the trigger logic as regex patterns inside the script and never mentions an API key for the hook; the 149-row Jev-vs-Haiku comparison is described as an included measurement, not something the hook runs. Verify current behavior in the repository before relying on it.

Question 7

What is the 149-row comparison mentioned in the description?

Accepted Answer

The GitHub description says the repository includes a measured 149-row comparison of TypeSafe Jev against Claude Haiku 4.5. The captured README excerpt does not show its results or methodology, so read the repository file itself rather than quoting numbers from the description.

TypeSafe Jev Tools

Highlights

Quickstart

Questions about this entry

Reactions & coverage

From the Jev guides

From the Jev AI model community