#guardrails

Your AI coding agent is only as reliable as the harness around it. Measure that harness in seconds with harness-score.

Fable-style spec + evidence gate for Claude Code + Codex. Makes Opus/Codex work under Fable-like discipline: blocks every edit until a deterministic spec passes, and there is no "done" without live acceptance evidence. Spec-first, verification-gated, forbidden-paths enforced.

fak — the Fused Agent Kernel: one Go binary that turns a tool-using agent (Claude Code, Codex, Cursor, any OpenAI/Anthropic/MCP client) into a managed agent: cache-stable model traffic, context compaction + crash resume, nanosecond tool-call policy, local GGUF serving with SSD expert offload.

Stateful agentic engineering workflow for local coding agents

EvalBot — local-first chatbot security & quality evaluation (FastAPI + Next.js). Evaluate chatbot answers against your own docs & guidelines with ML/NLP + AI-judge scoring. Apache-2.0.

A practical workbench of tools + methodology for solo devs driving an AI coding agent (Claude Code, Cursor, Copilot): safety hooks, secrets crypto, invariant + test-selection tooling, leak scanner, plus skill & memory system templates.

An audit gate for AI agent outbound messaging: hold the draft, let a senior agent review it (no private-data leak, prompt-injection resistance), then release. Reference implementation.

Building smart, scalable AI agent systems. Harness AI. Reclaim Time. Amplify Impact.

AI Guardian Firewall — a local, user-space, agent-agnostic firewall that mediates an autonomous AI agent's actions (files, shell, network, services) with a deterministic policy boundary, a tamper-evident audit log, and a human-in-the-loop approval cockpit. No kernel modules. Apache-2.0.

The authorization layer for AI agents — allow/deny/ask policy, human approvals, and tamper-evident receipts on every tool call. Zero-dep, framework-neutral, no SaaS.

The trust control plane for enterprise AI agents — real-time policy verdicts in under 800ms and litigation-grade evidence of every decision. Pharos decides. Pharos proves.

How far does this command reach? Guards AI coding agents against shell commands that leave your machine. Zero dependencies.

Execution substrate for governed agentic workloads; behavioral contracts, action budgets, pluggable memory, and portable skill bundles

Pragmatic AI Engineering Skills Library for LLMs, RAG, Agents, MCP & Spring AI

Same agent action, safe or dangerous by context: email team vs investor; delete scratch vs prod config. A safety layer routes each action: auto, confirm, escalate, block. A reading model auto-executes 0% of unsafe actions vs a baseline's 38% (McNemar p<0.001). Plus a runtime loop: an agent earns or loses autonomy with each counterparty over time.

AI/LLM automation reference: RAG, tool calling, human review, evals, observability, safety gates, and local deterministic samples.

Production-shaped agentic RAG: LangGraph plan->act->verify agent + hybrid retrieval + guardrails + an eval gate in CI. Provider-agnostic; runs offline with no keys.

Read-only first-pass incident triage on the Claude Agent SDK + MCP: classifies an alert, retrieves the matching runbook, and proposes a cited first response — or abstains to a human. Measured over four runs: 100% abstention with 0 missed escalations, ~$0.014 per incident.

Finance-owned FP&A tool where code computes every number and the LLM only explains — with a deterministic audit proving zero fabricated figures. Synthetic SaaS data.

AI AgenticSafe - stop your AI agent from breaking production at 3 a.m. 8 battle-tested doctrines, 3 real post-mortems, zero dependencies. By Yacine Mahboub, Founder of WEVIA.

Switchback Governance · 折返治理 — the human-in-the-loop governance layer for multi-agent teams (GOAI 2026 Track 1)

TypeScript CLI for installing agent skills, hooks, and CI/publish templates for Claude Code and related tools.

Do your agent pipeline's own checks catch planted faults? Measured on LangGraph, CrewAI and AutoGen: median 16.7%. One prompt-level change takes it to 55.0%. Pre-registered spec, Apache-2.0 harness, every raw trace published.

My Claude Code toolkit. Currently guard: hooks that scan agent-driven package installs before they run.

A guarded, transparent git for AI coding agents — PATH shim with per-agent worktree routing, deny guardrails, commit provenance, and audited bypass

A seatbelt for the AI acting on your systems — deterministic, open-source governance gate for AI-agent actions

Unified open-source security shield for agentic AI systems — defense-in-depth prompt-injection protection (canary tokens, agent-trace alignment audit, tool-call guarding, PII/secret scanning).

GitHub Action that lints CLAUDE.md-style agent constitution files for missing operational guardrails — spend limits, injection defense, escalation paths, secrets rules

Self-hosted, OpenAI-compatible AI gateway that splits your control plane (auth, guardrails, budgets, encrypted key vault, cost & observability) from the translation engine - swap Bifrost / LiteLLM / Portkey / Direct per provider at runtime. Own your keys; contain a compromised engine to one in-flight request.

Open-source trust layer for autonomous AI — gate agent, robot, smart home, and industrial actions before they run. Policies, HITL, Ollama/OpenAI, audit. MIT. npm @sanctum-runtime/sdk