jailbreak plugins

by claude-code-templates

Screens every prompt going into the conversation and every reply coming out of it with TypeSafe's Jev, a System One decision model, the way TypeSafe's guardrails cookbook does: one request per message asks a battery of hazard questions (jailbreak, harm or a crime, a diagnosis or a dosage, self-harm) and scores how much harm complying would do. The thresholds live here, in a named policy (strict or permissive), and turn the probabilities into pass, review (the user is asked), block (the prompt is dropped, or the reply withheld) or support. Falls back to the engine's built-in classifier when no key is set.

by AgentGuards

LLM security guardrails for OpenAI Codex — jailbreak detection, prompt-injection and web-content scanning, data-exfiltration blocking, and destructive-command authorization. Bundles enforcing hooks plus the AgentGuards MCP server.

by Ross Young

Agentic-Security: SAST + SCA + secrets + IaC + LLM-sec scanner with deterministic fix toolchain. Exposes a Model Context Protocol (MCP) server with 17 tools that Codex CLI can call directly.

by Matt Corbett

AI/LLM red-teaming team — agents (ai-redteam-lead, adversarial-testing-engineer) for the layer answering 'can this AI system be made to do harm, leak data, or exceed its authority — and how do we harden it?': threat modeling + rules of engagement, the attack taxonomy (OWASP LLM Top 10 2026 + MITRE ATLAS), direct vs indirect prompt injection, jailbreaks (roleplay/encoding/many-shot/crescendo), data exfiltration & training-data extraction, agentic tool-abuse / excessive agency, multimodal attacks, and defense-in-depth remediation. Fluent in automated red-team harnesses (PyRIT, Garak, Promptfoo red-team, Giskard) and likelihood×impact severity. skills, a knowledge bank (attack-taxonomy decision tree + 2026 patterns), and templates. Distinct from llm-evaluation-engineering (quality-regression eval), trust-and-safety (content-moderation / T&S policy), and security-engineering (app/infra pentest) — the adversarial AI-security layer over model- and agent-based systems. Requires ravenclaude-core@>=0.7.0.