Better Agents
Scaffold agent projects with testing, evaluation, prompt versioning, observability, and MCP-ready development conventions.
- Source repo
- langwatch/better-agents
- Stars
- ★ 1.6k
- Last updated
- 4mo ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 54/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformClaude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A developer starting an Agno project with Claude Code who wants Scenario end-to-end tests and versioned prompts from the first iteration.
- Not a fit if
- Teams wanting a ready runtime: it only scaffolds projects
- Teams avoiding a LangWatch cloud account and API key
- Developers not using Claude Code, Cursor, Antigravity, or Kilocode
- Source review
- 54/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
Better Agents is a CLI and set of development standards for initializing new agent projects. It is designed to work alongside a selected agent framework, including Agno, Mastra, and LangGraph, and supported coding assistants such as Claude Code, Cursor, Antigravity, and Kilocode CLI. Its generated project layout includes application code, Scenario end-to-end tests, evaluation notebooks, versioned prompts, a prompt registry, MCP configuration, and AGENTS.md guidance. The product produces and standardizes a local project foundation; it is not documented as a hosted agent runtime. It fits teams that want testing, prompt collaboration, evaluation, and observability conventions in place from project creation.
Running better-agents init . or better-agents init my-awesome-agent starts an interactive CLI flow for selecting a programming language, agent framework, coding assistant, LLM provider, and API keys. It creates project assets including app/ or src/, tests/scenarios/, tests/evaluations/, prompts/, prompts.json, .mcp.json, AGENTS.md, .env, and .gitignore. tests/scenarios/ is for Scenario end-to-end conversational tests, while tests/evaluations/ holds datasets and notebooks for evaluating pipeline tasks such as RAG and classification. Prompts are stored as versioned YAML files under prompts/ and controlled through prompts.json; .mcp.json configures MCP servers for the selected coding assistant.
- A developer starting an Agno project with Claude Code who wants Scenario end-to-end tests and versioned prompts from the first iteration.
- A Mastra team that needs YAML prompt files and
prompts.jsonto coordinate prompt work across collaborators. - An engineering team building a LangGraph workflow with RAG or classification steps that needs a place for evaluation datasets and notebooks.
- A technical lead standardizing new agent repositories around MCP configuration, observability, tests, and project conventions.
- A Cursor, Antigravity, or Kilocode CLI user who wants a consistent starting structure for a new agent codebase.
How do you install or deploy this agent?
Requirements are Node.js 22+, npm or pnpm, a supported coding assistant, a LangWatch API key, and an API key for the selected LLM provider. Install globally with npm install -g @langwatch/better-agents. Or run without a global installation: npx @langwatch/better-agents init my-agent-project.
How do you use this agent?
Initialize the current directory with better-agents init ., or create and initialize a new directory with better-agents init my-awesome-agent. Follow the CLI prompts to select the language, framework, coding assistant, LLM provider, and required API keys. Anonymous usage telemetry is enabled by default; set BETTER_AGENTS_TELEMETRY=0 to opt out.
What are this agent's strengths and limitations?
- Scaffolds Scenario end-to-end testing, evaluation notebooks, versioned prompts, and a prompt registry together rather than only creating application folders.
- Includes
.mcp.jsonso MCP configuration for the chosen framework and Scenario testing can be part of the new project. - The initialization flow supports choices of language, agent framework, coding assistant, and LLM provider.
- Separates behavioral tests, offline evaluation work, and prompt assets into explicit
tests/scenarios/,tests/evaluations/, andprompts/locations.
- Requires Node.js 22+ and npm or pnpm.
- Adoption requires both a LangWatch API key and an API key for the chosen LLM provider.
- The documented coding-assistant choices are limited to Claude Code, Cursor, Antigravity, and Kilocode CLI.
- It establishes project structure and guidance rather than supplying a deployed runtime; the actual framework implementation remains the project’s responsibility.
How does this agent compare with similar options?
Better Agents is complementary to Agno, Mastra, and LangGraph: those frameworks hold the agent implementation, while Better Agents scaffolds the surrounding testing, evaluation, prompt, and MCP conventions. It also works with, rather than replaces, coding assistants including Claude Code, Cursor, Antigravity, and Kilocode CLI.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Better Agents This agent | 54 · Major gaps | CLIFree + model costs | ★ 1.6k | 4mo ago | TypeScript | Claude Code |
| Rulesync | 69 · Some gaps | CLIFree | ★ 1.5k | today | TypeScript | Codex · Claude Code |
| zero2Agent: Zero-to-Production Agent Engineering Tutorials | 61 · Some gaps | Web appFree + model costs | ★ 541 | 4d ago | Python | — |
| Kite Agent Framework | 51 · Major gaps | CLIFree + model costs | ★ 11 | 2mo ago | Python | OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence: CLI guides user to select LLM provider and may request API keys, but least privilege and data flow are not clearly stated. Telemetry is enabled by default but can be disabled; collected data not specified. Dependencies have security overrides (e.g., picomatch, axios) but no vulnerability scan evidence. External effects include creating project files, installing dependencies, possibly calling external APIs, but user confirmation steps are not explicit. Rollback mechanism not mentioned. Source attribution: author is LangWatch, but not verified. Deductions: least privilege, user confirmation, data flow transparency, sensitive data handling, external effects, and rollback lack clear evidence.
Evidence: README and package.json descriptions are consistent; CI config includes typecheck, lint, build, unit tests, and E2E tests, but no test results provided. Dependencies are common and version-pinned, but availability not verified. Failure messages not documented. Deduction: failure messages lack concrete examples.
Evidence: Targets developers, supports multiple frameworks and coding assistants, scenarios clear. Capability boundaries described in docs (e.g., Node.js 22+, specific assistants). Trigger precision: CLI commands clear. Environment fit: supports multiple OS, but not explicitly stated. Deduction: environment fit details insufficient.
Evidence: Clear documentation structure (docs/), install instructions clear, naming stable (better-agents), examples and FAQ present (in docs), known limitations not explicitly listed, MIT license complete, CHANGELOG.md and versioning present, maintenance responsibility by LangWatch. Deduction: known limitations not explicitly listed.
Evidence: Output is a project scaffold, directly usable; value in standardization and best practices; cost is installing and using CLI. Deduction: no actual usage effectiveness or performance data.
Evidence: README claims (e.g., 'production-ready') lack specific evidence; no test results or independent verification. Deduction: claim traceability insufficient, lack of cross-source corroboration, facts and inferences not clearly separated.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Telemetry is enabled by default; data collection scope should be clarified.
- API key handling is not detailed; caution advised.
- No rollback mechanism; failed initialization may leave partial files.
FAQ
Does it deploy or run an agent for me?
app/ or src/.Which credentials are required?
Can telemetry be disabled?
BETTER_AGENTS_TELEMETRY=0; anonymous usage telemetry is enabled by default.