guardrails 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 Aria Han

KERNEL is a Claude Code and Codex runtime for direct, decomposed execution with local memory and minimal hooks. Version 9.11.0 deploys the simplify deletion doctrine, preflight and verifier contract. Skills invoke as /kernel:<name> on Claude Code and $kernel:<name> on Codex; run /kernel:help first. Agent-readable summary: llms.txt.

by Yiming Wu

Structured guardrails for AI coding: brainstorm, plan, execute, review, and verify.

by Hashgraph Online

Install and manage the local-first HOL Guard package for Claude Code. This plugin guides setup and status checks; HOL Guard itself installs and owns runtime security hooks for risky commands, secrets, packages, MCP, and supply-chain activity.

by Nizar Selander

TDD enforcement and guardrails for Claude Code, Codex, and GitHub Copilot CLI.

by Clear-Sights

Holds Claude Code to its word — every claim checked against the agent's own logged record; faked checks blocked, not warned (sixteen narrow advisory self-checks excepted). Two-tier catalog (14 pre-checks, 30 Stop gates), each shipped at measured zero false positives.

by Yasyf Mohamedali

Skills for the capt-hook declarative hook framework: bootstrap gates and nudges from a repo's conventions, and translate style guides into enforced StyleRules.

by Dwarves Foundation

Spec-driven workflow with shared Claude Code and Codex guardrails.

by smrafiz

Shell-level guardrails for Claude Code. Command blocking, code security scanner, self-teaching, token economy, agent routing, MCP profiles, and session memory.

by Matt Corbett

LLM-evaluation engineering team — agents (eval-strategy-lead, eval-harness-engineer) for the question every team shipping an AI feature has to answer: is it getting better or worse? Eval design (task-grounded metrics, offline-vs-online split, sample size, a ship-gate decided as a number before the run), and the harness that proves it (frozen golden sets with provenance, LLM-as-judge rubrics with a bias audit and human calibration, CI regression gates, guardrail/red-team suites). skills, a decision-tree knowledge bank (an eval-method Mermaid tree + a dated 2026 tooling/method map), best-practices. Engineering decision-support, not a safety certification; model/judge/tooling specifics are volatile — retrieval-dated + verify-at-use; no eval-data PII. Seams: retrieval quality -> ai-rag-engineering; model selection -> ai-coding-model-guidance; classical-ML training/metrics -> ml-engineering; product A/B -> experimentation-growth-engineering. Requires ravenclaude-core@>=0.7.0.

by Matt Corbett

Prompt & context engineering team — agents (prompt-architect, prompt-implementation-engineer, prompt-reliability-engineer) that own the layer between the model and the app: prompt & context strategy (task decomposition, pattern selection across few-shot / CoT / decomposition / role, system-vs-user structure), the output-format contract (JSON mode, tool/function-calling schemas, few-shot curation, token budgeting), and prompt reliability (eval sets & regression, prompt versioning/CI, prompt-injection & jailbreak defense at the prompt layer). Provider-neutral. skills, a knowledge bank with Mermaid decision trees + a dated 2026 reference, best-practices, templates. Seams: which model → ai-coding-model-guidance / claude-api; retrieval → ai-rag-engineering; eval at scale → llm-evaluation-engineering; adversarial testing → ai-red-teaming; the app → claude-app-engineering / backend-engineering. Requires ravenclaude-core@>=0.7.0.

by Karan Bansal

A one-file PreToolUse door-guard that bounces destructive Bash commands (rm -rf, DROP TABLE, curl|sh, force-push, device wipes) before they run. Blocks 45/45 known footguns, 0 false positives.

by Diego Romero

Autopilot for spec-driven development. The human signs off and merges; planwright flies the spec. Scales to a multi-spec fleet: an agentic software factory on Claude Code primitives.

by Arockiaraj Rayappan

Your team's coding standards, in context. Connects to a dev-playbook server (run it locally with Docker) over MCP to serve guardrails, definition of done, per-language rules and task workflows - and scaffolds them for a repo that has none. Run /dev-playbook-init to set it up.

by Dtwo

Manage Dtwo gateways, policies, and Rego with the Dtwo MCP server.

by Dongmin Yu

Seven defensive hooks: blocks recursive rm on $HOME, download-and-execute pipelines, secrets-file access, path-less formatter runs, credentials in the staged diff, and the two zsh quoting mistakes that fail silently; warns on stale lockfiles and shellcheck findings.

by Kleber Maia

Make your coding agent work like a staff engineer: interview before building, preview before tests, gated lifecycle, plain-language handoffs. Stack-agnostic, zero dependencies.

by idnotbe

Hook-based security guardrails for Claude Code's --dangerously-skip-permissions mode. Blocks destructive commands, protects secrets, and auto-commits safety checkpoints.

by Ponlawat

Knowledge graph of feature business logic and code fences, stored as an Obsidian vault in each project.

by Jev-decision

Build AI-powered software with TypeSafe System One and Jev using typed judgments, calibrated probabilities, and composable decision primitives.

by Jev-decision

Build AI-powered software with TypeSafe System One and Jev using typed judgments, calibrated probabilities, and composable decision primitives.

Let an AI coding agent work a whole ticket without watching every command it runs.

by ojiman

A small personal collection of Agent Skills, each written up from something that actually went wrong on a real project.