Starter kit for building test-first, easily deployable AI agents.
- Scaffolded Poetry project with CLI and stub agent
- Added Pytest plugin, fixtures, and green sample test
- Configured GitHub Actions
ci.ymlto run test suite - Reserved package names by publishing
mini-agent-harness==0.0.0to PyPI and NPM - Flesh out
README.mdwith badges, quick-start guide, and differentiators - Add GitHub Actions release workflow to publish to PyPI & NPM on version tags
- Bump working version to
0.1.0.dev0(Python) and0.1.0-dev(NPM) - YAML tool registry with auto-import & CLI stub generation
- Implement core ReAct-style agent loop (
mini_agent_harness.core.loop)
(nothing at the moment)
-
mini-agent serveFastAPI server with streaming SSE endpoint - Streaming LLM responses (chunked) for realtime UI
- Prompt templating with Jinja2 and basic conversation memory
- React webapp (
webapp/) with shadcn/ui chat + playground - Testing harness Phase 1: DeepEval metrics and Guardrails schema checks
- Robust parsing / error handling for ACTION/ARG/FINAL blocks
- One-click Vercel deploy (
deploy/vercel.json) and README badge - Cloudflare Workers edge target via
wrangler deploy
- Testing harness Phase 2: LangSmith golden conversation replay
- Docs site and launch assets (logo, screenshots, blog post)
Phased approach (adapted from 6-week outline):
- Core agent loop & expanded CLI
- Testing harness Phase 1 (unit metrics)
- Web UI skeleton
- One-click deploy targets (Vercel → CF Workers)
- Documentation polish & examples
mini_agent_harness/cli.py– entry-point formini-agentCLI ✅mini_agent_harness/core/__init__.py– stubAgent+AgentResult✅mini_agent_harness/testing/– pytest plugin & fixtures ✅.github/workflows/ci.yml– CI running test suite ✅README.md– currently being updated