User repository on GitHub
TypeSafe AI driven agent tool execution guard for Claude Code hooks
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Agents & MCP
Agent Guard
Claude Code PreToolUse hook that screens Bash, file edits, and tool calls with Jev risk, policy, and allow-ask-deny judgments before execution.
- Stars
- 2
- Language
- Python
- License
- MIT
- Published by
- Community
- Author
- NieXi
- Last verified
- Oct 2026
Highlights
- Returns native allow, ask, or deny hook output for matched Bash, Write, Edit, and tool calls.
- Combines a graded destructive-risk score, a policy-violation probability, and a final execution choice.
- Installs via package managers with setup guidance and per-repo hook configuration.
- Supports offline mock checks, colored terminal reports, and JSONL audit logging.
Watch out
Needs a TypeSafe API key for live judgments; offline mode is for testing only and config mistakes fail closed. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
Yes. The digest states deny applies in every permission mode, including bypass modes, so switching modes cannot bypass it. Ask decisions in headless no-prompt sessions convert to deny instead of silent approval.
For live review, yes. Configure the key in the global or project config file or as an environment variable; the offline mock flag exists only for fast testing and demos without calling the API.
Use the check subcommand with a command string or tool payload, optionally with the mock flag. It prints the allow, ask, or deny outcome with reasons before you wire the hook into settings.
It fails safe: missing keys or malformed responses trigger a protective fallback rather than silent bypass, and an audit log variable can persist every reviewed call as JSONL for review.
A destructive_risk Score on a four-level scale from read-only to irreversible, a policy_violation Noul for whether a safety rule is crossed, and a decision Choice of allow, ask, or deny.
Run agent-guard check --cmd "git status" for a read-only command, or add --mock to simulate without an API key. You can also pipe a PreToolUse JSON payload into agent-guard hook.
In a global ~/.agentguardrc file, a project-local ./.agentguardrc, or the TYPESAFE_API_KEY environment variable. The config file is recommended to avoid losing the variable in subprocesses.
More Jev resources like this
1GitHub stars
Demo of the LangChain Jev harness pattern in LangGraph: a fast classifier routes events and gates a risky tool call, leaving open-ended reasoning to a chat LLM. Entry is POST /events; output lands in JSONL files.
Jev Agents & MCP#community#github
1GitHub stars
Offline prototype replaying public Shin 2017 EEG sample-by-sample to trigger intents, gate actions with a Jev policy question, and confirm via repeat or imagery before a mock agent acts.
Jev Agents & MCP#community#github
1GitHub stars
Claude Code plugin that puts Jev judgments in the agent loop to route tool calls, verify outputs, judge completion, and pick next actions with metered calls.
Jev Agents & MCP#community#github
Community
pur4v/jeveloperPython
pur4v/jeveloperPython1GitHub stars
Python guardrail that intercepts agent tool calls, sends agent state plus tool details to Jev, and allows, blocks, or reroutes execution based on Choice, Score, and Noul answers.
Jev Agents & MCP#community#github
0GitHub stars
Browser-use agent that completes flight searches in real Chrome: Jev picks each next action from an indexed DOM table while a small model handles only text entry.
Jev Agents & MCP#community#github
1GitHub stars
Backoffice email pilot where the LLM generates and uses CRM and ERP tools while Jev triages, gates tool calls, and verifies drafts before sending, with Telegram human approval.
Jev Agents & MCP#community#github
Community
lucianoon/backoffice-agentsPython
lucianoon/backoffice-agentsPythonFrom the Jev community
Posts from builders shipping with Jev right now.