Security & Guardrails

Jev judges prompt injection or unsafe content; local rules forward, flag or block the request.

↳ Adds an inspectable content check ahead of an existing HTTP service.

Jev scores destructive actions, scope and output risks; configuration chooses warnings or confirmation.

↳ Defaults to shadow mode and fails open on API errors; it is not a standalone security boundary.

Local rules handle explicit denies and allows before Jev assesses authorization and risk for bash, write, and edit.

↳ Keeps policy and decision records available for adjusting operation gates.

Uses conversation and command results to assess whether to ask the user, whether an action exceeds scope, and whether to retry.

↳ Makes checkpoints that may require user confirmation explicit.

Jev evaluates tool inputs or outputs in session context; local rules allow, warn or block.

↳ Adds review points around Agent tools, with behavior depending on the host client.

Jev returns APPROVE, DENY or ESCALATE, with local thresholds and policy determining handling.

↳ Demonstrates an approval integration; the author labels it a proof of concept, not production assurance.

Deterministic rules run first; Jev assesses remaining call risks and tool relevance for threshold-based handling.

↳ Approval-call failures request human review; model judgments cannot guarantee detection of every risk.

Jev scores jailbreaks, harmful content and secret disclosure; thresholds or evaluation errors reject requests.

↳ Demonstrates gateway integration, not guaranteed prevention of all attacks.

Assesses inputs such as test names, documentation claims and supporting evidence.

↳ Keeps model advice distinct from deterministic rules and records checks.

Uses Jev to judge rule violations, irreversible actions, and task drift, feeding results back to the Agent.

↳ Brings rule findings and actions needing confirmation into the active workflow.

Evaluates candidate page elements before the extension applies local hiding rules.

↳ Turns page judgments into reusable rules.

Sends proposed actions and permission rules to Jev, then uses local logic to allow, deny, or ask the user.

↳ Connects permission policies to judgments about individual tool actions.

Classifies text that passes local gates; local mapping decides whether to allow or suppress it.

↳ Makes OTP handling and fail-open behavior explicit in code.

Maps attack conditions and impacts to discrete metrics; local CVSS formulas calculate the numeric score.

↳ Exposes the selected metrics and vector alongside the final score.

Assesses messages or recent member history; the bot applies deletion, warnings or timeouts.

↳ Connects judgments, moderation logs and manual correction controls.

Adds a destructive-risk judgment after local shell checks and refuses commands above the configured threshold.

↳ API failures generally allow the command with a warning, so it is not complete system protection.

Sends scan rules, matched snippets, and file context to Jev to assess likely true positives and suggested handling.

↳ Organizes rule-based scan findings into a queue for human review.

Judges risks such as exposed credentials or destructive changes; local rules warn or block a commit.

↳ Presents risk labels with the corresponding diff before commit.

Feeds file contents and security criteria into Jev to assess risks such as data exfiltration, obfuscated loaders, and credential leakage.

↳ Provides rapid pre-execution safety screening across multiple file formats without needing a heavy static analysis toolchain.

Choice clean / suspicious_pua / malicious, Score 0–4, and Noul questions for packing, C2, persistence, and injection; local policy maps the answers.

↳ Turns heuristic file features into thresholded structured answers rather than a prose verdict; this site has not run or validated protection.

For each text chunk, asks twelve PII-category Nouls and a three-level sensitivity Score.

↳ Judgment runs on TypeSafe servers; local code only locates spans, so this is not an offline privacy scanner.

Before each tool call, Jev receives the user's request (or pinned task), the proposed command, the agent's explanation, and recent conversation. It answers a Choice for intent (on task / off task / injected / needs more context) and a Score for risk (safe / needs approval / harmful). Local code turns these into run, ask the user, or warn. If Jev says more context would help, the extension adds it and asks again.

↳ Adds structured choices or scores to the workflow; performance and cost benefits have not been independently verified.

Submits code snippets to Jev to evaluate hardcoded credential risk, returning boolean and confidence scores.

↳ Evaluates the feasibility and precision of low-latency non-generative models in pre-commit security gates.