Scores each action an agent wants to take against the job it was given: allow, ask a person, or block.
@shaporJev Sentinel reads an action an agent is about to take and scores it against the task that agent was given. The result is allow, ask a person, or block, with a short reason.
A check command can show a block for reading a private key when the job is fixing a login form.
It was built at the San Francisco Jev hackathon.
Stops a coding agent from finishing on a claim: it hooks into Claude Code and Codex, keeps a ledger of what the agent really did, and asks Jev only for the judgment calls.
Jev
TypeScript
Node.js
Claude Code
CodexAgent skills and a coding harness where Jev picks the relevant files, so your coding model only reads what matters.
Jev
PythonCodexClaude CodeA Claude Code hook that asks Jev about every shell command and blocks the ones that would delete something that matters.
JevPythonClaude CodeA Python experiment where Jev picks each actor's next legal move, and the simulation applies it.
JevPythonAsk a repository's history a plain question and it points at the commit that did it.
JevPythonA compiler driver that asks Jev whether to inline each function call, aiming to beat clang -Oz on binary size.
JevPython
LLVMClang
CodeRabbit Top10
