Projects.

Durable goals for Claude Code and Codex.

Project-scoped /goal state, audit-gated completion, statusline visibility, MCP tools, and Claude Code ↔ Codex relay for long-running coding tasks.

Toolchain for the Agent Skills ecosystem.

Lint, score, and route SKILL.md files. Parse, validate, and quality-score Agent Skills per the agentskills.io spec. BM25 routing, zero LLM dependencies.

Open protocol for AI agents to control web browsers.

TypeScript and Python SDKs with native MCP integration. Semantic selectors built on the accessibility tree.

Deterministic Browser Agent Runtime.

Replayable, verifiable browser executions via CDP virtual time, network record/replay, and portable determinism capsules.

Universal Semantic Element ID.

Stable cross-run element identity for browser agents. Fuses ARIA semantics and DOM structure into signatures that survive UI churn.

YAML spec and Python SDK for production agent reliability.

Cost control, tool-use security, and audit trails enforced at the runtime boundary instead of trusted to the prompt.

Bridge between AI coding agents and Slack.

Wire Claude Code, Codex CLI, and other agents into a Slack channel. Simple webhooks, zero config.

Data augmentation for automatic speech recognition.

PyTorch implementation of SpecAugment, a simple data augmentation method for ASR.

Cluster-GCN for large graph training.

PyTorch implementation of Cluster-GCN, an efficient algorithm for training deep and large graph convolutional networks.

End-to-end text-independent speaker verification.

Identify speakers from short audio clips without depending on what they're saying.

Speech translation API powered by Whisper and GPT.

Combines OpenAI Whisper and GPT-3.5 to deliver accurate, fluent translations from spoken audio.