# Agent harnesses and coding agents

> 10 agent harnesses ranked by the Anchor benchmark. Leader goose (BB). Finished programs that run the agent loop for a person or a pipeline, in a terminal, an editor or the vendor's cloud. They plan, call tools, edit files and run commands, where a framework is a library you build that loop with. Compared on what they ask before acting, what the sandbox and the network allow by default, MCP support, headless use and the telemetry they send.

- Canonical: https://www.anchorterminal.com/categories/agent-harnesses
- Markdown: https://www.anchorterminal.com/categories/agent-harnesses.md (~2,600 tokens)
- Slim: https://www.anchorterminal.com/categories/agent-harnesses.min.md (~480 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/categories/agent-harnesses.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (full text: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- Updated: 2026-10-05

Finished programs that run the agent loop for a person or a pipeline, in a terminal, an editor or the vendor's cloud. They plan, call tools, edit files and run commands, where a framework is a library you build that loop with. Compared on what they ask before acting, what the sandbox and the network allow by default, MCP support, headless use and the telemetry they send.

- Tools ranked: 10 · agent-ready (BB or better): 4 · accept x402: 0 · hosted endpoints: 0 · desk reviews by the panel: 18
- JSON: https://www.anchorterminal.com/api/v1/tools.json (list) · https://www.anchorterminal.com/api/v1/rankings.json (ranked) · https://www.anchorterminal.com/api/v1/x402.json (payable) · https://www.anchorterminal.com/api/v1/capabilities.json (by capability)
- Grades run AA, A, BB, B, C, D, E, F · methodology: https://www.anchorterminal.com/benchmark/

- Capabilities in this category: agent.harness, agent.mcp-client, agent.multi-agent
- https://letme.dev/agent.harness picks the top-graded tool in this list and says how to call it direct; calling through letme comes later (https://www.anchorterminal.com/letme/index.md)

## Ranking

| # | Tool | Vendor | Kind | Category | Grade | Score | Confidence | x402 | Auth | Where | Reviews | Page |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 52 | goose | Agentic AI Foundation (originally Block) | Agent harness | Harnesses | BB | 73.9 | medium | no | None | local | 2.5/5 (2) | https://www.anchorterminal.com/tools/goose.md |
| 58 | OpenAI Codex | OpenAI | Agent harness | Harnesses | BB | 73.4 | medium | no | OAuth or key | local | 3/5 (2) | https://www.anchorterminal.com/tools/openai-codex.md |
| 72 | Gemini CLI | Google | Agent harness | Harnesses | BB | 72.3 | medium | no | OAuth or key | local | 3.5/5 (2) | https://www.anchorterminal.com/tools/gemini-cli.md |
| 92 | OpenHands | All Hands AI | Agent harness | Harnesses | BB | 70.9 | medium | no | OAuth or key | local | 2.5/5 (2) | https://www.anchorterminal.com/tools/openhands.md |
| 134 | OpenCode | Anomaly | Agent harness | Harnesses | B | 68 | medium | no | None | local | 2/5 (2) | https://www.anchorterminal.com/tools/opencode.md |
| 222 | Claude Code | Anthropic | Agent harness | Harnesses | B | 62.2 | medium | no | OAuth or key | local | none | https://www.anchorterminal.com/tools/claude-code.md |
| 239 | Cline | Cline Bot Inc. | Agent harness | Harnesses | C | 60.8 | medium | no | OAuth or key | local | 2/5 (2) | https://www.anchorterminal.com/tools/cline.md |
| 286 | GitHub Copilot CLI | GitHub | Agent harness | Harnesses | C | 57.9 | medium | no | OAuth or key | local | 2.5/5 (2) | https://www.anchorterminal.com/tools/github-copilot-cli.md |
| 385 | Aider | Aider AI LLC | Agent harness | Harnesses | D | 47.1 | high | no | None | local | 1/5 (2) | https://www.anchorterminal.com/tools/aider.md |
| 441 | Cursor CLI | Cursor | Agent harness | Harnesses | F | 35.8 | low | no | OAuth or key | local | 1.5/5 (2) | https://www.anchorterminal.com/tools/cursor-cli.md |

Scores are from public evidence against the published checklist (https://www.anchorterminal.com/benchmark/), with Performance and Task success pending. p95 latency and context cost come from our probes, which haven't run yet.

## Summaries

### 52. goose, BB (73.9)

Open-source general-purpose agent written in Rust, with a desktop app, a CLI and an embeddable server. Telemetry off until the user opts in, with the collected fields listed. Autonomous mode, which approves every tool call, is the default.

- Page: https://www.anchorterminal.com/tools/goose · Markdown: https://www.anchorterminal.com/tools/goose.md · JSON: https://www.anchorterminal.com/api/v1/tools/goose.json
- Capabilities: agent.harness, agent.mcp-client, agent.multi-agent

### 58. OpenAI Codex, BB (73.4)

OpenAI's coding agent for software development tasks. Sandbox on by default on macOS, Linux and Windows, with the network off and `.git` and `.codex` read-only. Pre-1.0 at 0.160.0, with a minor every few days and no breaking-change section in release notes.

- Page: https://www.anchorterminal.com/tools/openai-codex · Markdown: https://www.anchorterminal.com/tools/openai-codex.md · JSON: https://www.anchorterminal.com/api/v1/tools/openai-codex.json
- Capabilities: agent.harness, agent.mcp-client

### 72. Gemini CLI, BB (72.3)

Google's open-source coding agent for the terminal, in TypeScript on Node 20 or newer. Apache-2.0, CI passing on main, and 583 open issues with priority labels. Sandboxing is off by default, and the default macOS profile allows network.

- Page: https://www.anchorterminal.com/tools/gemini-cli · Markdown: https://www.anchorterminal.com/tools/gemini-cli.md · JSON: https://www.anchorterminal.com/api/v1/tools/gemini-cli.json
- Capabilities: agent.harness, agent.mcp-client, agent.multi-agent

### 92. OpenHands, BB (70.9)

Open-source coding agent with a self-hosted web interface, local and remote execution, and scheduled or webhook-driven automation. A Docker container per conversation with `OH_CONVERSATION_RUNTIME=docker`, each with its own Agent Server. Confirmation mode is off by default in Agent Canvas, and the npm install gives the agent the host's whole filesystem.

- Page: https://www.anchorterminal.com/tools/openhands · Markdown: https://www.anchorterminal.com/tools/openhands.md · JSON: https://www.anchorterminal.com/api/v1/tools/openhands.json
- Capabilities: agent.harness, agent.mcp-client, agent.multi-agent

### 134. OpenCode, B (68)

Open-source terminal coding agent from Anomaly Innovations, with a TUI, a desktop app in beta, IDE and ACP integration, and a headless HTTP server with an OpenAPI spec and a TypeScript SDK. Runs with no key or account on free OpenCode Zen models. Most permissions default to allow, and SECURITY.md says the permission system is not a sandbox.

- Page: https://www.anchorterminal.com/tools/opencode · Markdown: https://www.anchorterminal.com/tools/opencode.md · JSON: https://www.anchorterminal.com/api/v1/tools/opencode.json
- Capabilities: agent.harness, agent.mcp-client, agent.multi-agent

### 222. Claude Code, B (62.2)

Anthropic's coding agent as a terminal program, also in VS Code, JetBrains, the desktop app and Anthropic-hosted cloud sessions. Six permission modes, allow, ask and deny rules down to command arguments, PreToolUse hooks, and managed settings that can disable bypass and auto mode. The sandbox is off by default and native Windows has none.

- Page: https://www.anchorterminal.com/tools/claude-code · Markdown: https://www.anchorterminal.com/tools/claude-code.md · JSON: https://www.anchorterminal.com/api/v1/tools/claude-code.json
- Capabilities: agent.harness, agent.mcp-client, agent.multi-agent

### 239. Cline, C (60.8)

Open-source coding agent that runs as a VS Code extension, a JetBrains plugin, a CLI and a desktop app, all on one TypeScript SDK since extension 4.0.0 (26 June 2026). Approval before edits and commands in the IDE, with command auto-approval off by default since 4.0.0. The CLI approves every tool by default outside ACP mode and starts on Cline's own provider.

- Page: https://www.anchorterminal.com/tools/cline · Markdown: https://www.anchorterminal.com/tools/cline.md · JSON: https://www.anchorterminal.com/api/v1/tools/cline.json
- Capabilities: agent.harness, agent.mcp-client, agent.multi-agent

### 286. GitHub Copilot CLI, C (57.9)

GitHub's coding agent for the terminal, built on the same agent harness as Copilot cloud agent (formerly Copilot coding agent), which works in GitHub Actions and opens pull requests. Asks before the first use of each modifying tool, and `--deny-tool` beats `--allow-all-tools` and `--allow-tool`. Free, Pro, Pro+ and Max interactions train GitHub's models by default since 24 April 2026.

- Page: https://www.anchorterminal.com/tools/github-copilot-cli · Markdown: https://www.anchorterminal.com/tools/github-copilot-cli.md · JSON: https://www.anchorterminal.com/api/v1/tools/github-copilot-cli.json
- Capabilities: agent.harness, agent.mcp-client, agent.multi-agent

### 385. Aider, D (47.1)

Terminal pair-programming tool that edits files in a local git repository through text edit formats rather than tool calls, builds a repo map with tree-sitter, and commits each change. Analytics opt-in and offered to 10 per cent of users, with a permanent opt-out and a local log. No release since 0.86.2 on 12 February 2026 and no commit since 22 May 2026.

- Page: https://www.anchorterminal.com/tools/aider · Markdown: https://www.anchorterminal.com/tools/aider.md · JSON: https://www.anchorterminal.com/api/v1/tools/aider.json
- Capabilities: agent.harness

### 441. Cursor CLI, F (35.8)

Cursor's coding agent in the terminal, run as `agent` (also `cursor-agent`). Allow and deny rules for shell, reads, writes, web fetches and MCP tools, with deny taking precedence. No CLI changelog, and versions are dates.

- Page: https://www.anchorterminal.com/tools/cursor-cli · Markdown: https://www.anchorterminal.com/tools/cursor-cli.md · JSON: https://www.anchorterminal.com/api/v1/tools/cursor-cli.json
- Capabilities: agent.harness, agent.mcp-client

## How we test this category

The same small repository task run headless in each harness with one MCP server attached, first fixing a failing test, then a task that needs the network. We check what it asks before acting, what the sandbox blocks, whether the run stops on its own, what it costs and what leaves the machine. This test hasn't run yet, so Task success is pending and the grades here come from the categories assessed from public evidence.