Question 1

Does this actually block the agent, or is it just another line of prompt to remember?

Accepted Answer

It runs in lifecycle hooks outside the agent's turn: PreToolUse before every tool call, Stop before the final response. The tool call does not go through until the verifier approves it, so the agent cannot simply decide a rule does not apply. The judge is also a separate typed model call rather than the same agent re-reading its own output in the same context that produced it.

Question 2

Can the agent turn verification off to get past a block it just hit?

Accepted Answer

Pause and resume are user-triggered only. /jevguard:pause and /jevguard:resume flip a flag file rather than an env var, since each hook is a fresh process, and the skill will not run on the agent's own initiative. The agent can create and update rules with the Write/Edit tools it already has and confirm the result with jevguard-validate, but it cannot reach for pause itself.

Question 3

What happens if the Jev backend is down or I forget to set the API key?

Accepted Answer

Verification fails open. With no key or URL set, the base URL defaults to https://api.typesafe.ai and the request goes out with no Authorization header, which works for a local backend that does not require one and returns 401 from the real API. A 401 counts as a failed verifier, so the hook stops checking instead of blocking you. Convenient during setup, and also a way for enforcement to disappear without an obvious signal.

Question 4

Does batching ten rules really cost about the same as checking one?

Accepted Answer

That is the project's claim, not a measurement the README provides. It argues cost by construction: multiple questions about the same content go into a single request following Jev's speculative fan-out pattern, so checking ten rules together is close to checking one. No latency or price numbers are published, so treat the ordering as a design rationale rather than a benchmark result.

Question 5

How do I install JevGuard?

Accepted Answer

For Claude Code, add the marketplace with /plugin marketplace add Jhonnyr97/JevGuard then /plugin install jevguard@jevguard and run /plugin configure to set the key and backend URL; Codex uses codex plugin marketplace add and codex plugin add.

Question 6

How are JevGuard rules created and where do they live?

Accepted Answer

/jevguard:init reads CLAUDE.md or AGENTS.md, the README, code conventions and Claude Code auto-memory to generate rules, /jevguard:rule adds one in plain language, and rules live in the hand-editable .jevguard/rules.json.

Question 7

Where does JevGuard keep the API key?

Accepted Answer

Never in the committed .jevguard/config.json. On Claude Code the key is held in the OS keychain via plugin userConfig, and elsewhere it comes from the JEVGUARD_API_KEY and JEVGUARD_BASE_URL environment variables.

JevGuard

About this project

Highlights

Questions about this entry

More Jev resources like this

From the Jev community