I build open-source infrastructure for reliable AI agents. The focus is on making agent behavior predictable: controlling context, running deterministic work without an LLM in the loop, enforcing what tools are allowed to do, and checking generated code before it ships. The repos below are independent but designed to fit together.

Pick the repo that matches the problem you have:

• My agent has too many tools or oversized tool outputs → contextweaver compiles large tool catalogs into bounded choices and firewalls big tool results to keep prompts within budget.
• My agent keeps repeating the same tool sequence → ChainWeaver compiles those repeated paths into typed, deterministic flows so the LLM is not re-invoked between steps that never change.
• I want to control what an agent's tool calls are allowed to do → AgentFence (a local policy firewall you run in front of MCP tool calls) or agent-kernel (the same kind of enforcement as an embeddable capability/policy layer inside your own runtime).
• I want to catch risky AI-generated code before merge → VibeGuard is an offline pre-merge gate that flags common security risks and AI-generation artifacts in a diff.
• I want to evaluate a decision policy on logged data before rolling it out → skdr-eval estimates how a candidate recommender, routing, or targeting policy would perform offline, with diagnostics on whether the estimate can be trusted.

• VibeGuard — fast, offline pre-merge check that flags common security risks and AI-generation artifacts in code diffs.
• ChainWeaver — compiles repeated, deterministic tool sequences into auditable typed flows, removing unnecessary LLM calls between steps.
• contextweaver — context gateway for tool-heavy agents that routes large tool catalogs to bounded choices and trims oversized tool results to control prompt tokens.
• AgentFence — local MCP policy firewall that evaluates each tool call and allows, denies, or asks for approval, with no cloud dependency or telemetry.
• agent-kernel — embeddable capability-based authorization layer that issues revocable, principal-scoped tokens and keeps a tamper-evident audit of what ran.
• weaver-spec — language-agnostic contracts and shared vocabulary so these components can interoperate without adopting all of them.
• skdr-eval — offline policy evaluation library (applied ML side project) for estimating policy performance from logged decisions before an A/B test.

• AgentFence and agent-kernel apply the same idea — deciding whether a tool call is allowed — at different integration points. AgentFence is a standalone local proxy you put in front of MCP tool traffic; agent-kernel is the embeddable library you call from inside your own agent runtime.
• contextweaver, ChainWeaver, agent-kernel, and weaver-spec are meant to compose: ChainWeaver handles deterministic execution, contextweaver controls context and token budget, agent-kernel enforces authorization, and weaver-spec defines the shared contracts that let them work together. Each can also be used on its own.

Open to feedback, issues, design discussion, and collaboration.