Skillget

harness

mcp
Guvenlik Denetimi
Basarisiz
Health Uyari
  • License — License: Apache-2.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Basarisiz
  • execSync — Synchronous shell command execution in .github/workflows/ci.yml
  • fs module — File system access in .github/workflows/ci.yml
  • process.env — Environment variable access in examples/living-wiki-jaeger/src/backend/env.ts
  • process.env — Environment variable access in examples/living-wiki-jaeger/src/backend/harness.ts
  • process.env — Environment variable access in examples/living-wiki-jaeger/src/backend/telemetry.ts
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

AI Harness build on top of typescript

README.md

PURISTA Agent Harness

Self-hosted TypeScript infrastructure for building provider-neutral LLM agent
systems inside your own application or platform.

The harness gives PURISTA applications a typed runtime boundary for:

  • direct agent invocation through session.agents;
  • workflow orchestration through session.workflows;
  • provider-neutral text, structured object, multimodal, embedding, and rerank
    model operations;
  • TypeScript, built-in, and MCP tools;
  • reusable skills;
  • state, sandboxing, durable workspace replay, logs, traces, and run events;
  • provider-neutral eval helpers for deterministic scorer tests and prompt
    candidate comparison;
  • provider adapters for OpenAI, Anthropic, Amazon Bedrock, and Azure AI Foundry.

This repository is not a SaaS product. It is lower-level infrastructure that
application teams embed in services, workers, CLIs, or local tools.

Mental Model

An agent is a typed LLM conversation loop. It prepares messages, calls a
model, executes tool invocations, feeds tool results back into the model,
validates the final output, and emits run events.

A workflow is application orchestration around one or more agent
invocations. Workflows sequence or parallelize agents, add deterministic logic,
request human approval, write durable state, and create artifacts.

Quick Start

npm install
cp .env.example .env
npm run build
npm test

Set OPENAI_API_KEY in .env; examples default to OPENAI_MODEL=gpt-5-mini.

The examples/quickstart workspace is the recommended starting point.
examples/showcase demonstrates mounted skills, custom TypeScript tools, and
multiple workflows with the OpenAI adapter. examples/living-wiki-jaeger
demonstrates a local file-backed research workspace with direct typed agent
invocation, optional typed workflows, SSE run observation, review gates,
artifacts, MCP, and Jaeger tracing.

See docs/getting-started/quickstart.md for the full walkthrough.

Verification

npm run lint
npm run typecheck
npm test
npm run test:coverage
npm run build

Documentation

Project Structure

  • packages/harness/ — Core runtime, contracts, ports, builder, sessions, tools, sandbox, telemetry, and test helpers.
  • packages/harness-openai/ — OpenAI model provider adapter.
  • packages/harness-anthropic/ — Anthropic model provider adapter.
  • packages/harness-bedrock/ — Amazon Bedrock model provider adapter.
  • packages/harness-azure-foundry/ — Azure AI Foundry model provider adapter.
  • examples/quickstart/ — Smallest runnable PURISTA harness example.
  • examples/showcase/ — Skills, TypeScript tools, and multiple workflow examples.
  • examples/living-wiki-jaeger/ — Local research workspace with Hono, React/Vite, SSE, artifacts, MCP, and Jaeger.
  • docs/ — End-user and operator documentation.
  • specs/ — Requirements and design contracts for implementation work.

Yorumlar (0)

Sonuc bulunamadi