APPLICATION

Claude Code: the agentic coding tool that lives in the terminal, and the reference harness for Anthropic models

claude-code

Claude Code is the agentic coding tool from Anthropic, and its own definition is deliberately modest: it lives in your terminal, understands your codebase, and executes routine tasks, explains complex code and handles git workflows through natural language. The official docs name three surfaces - terminal, IDE, and tagging @claude on GitHub. anthropics/claude-code stood at 148,438 stars / 24,902 forks on 2026-09-28, among the highest in this class; stated plainly, the repository has no LICENSE file (the GitHub API returns null), so it is publicly readable but not openly licensed - a material difference from the Apache-2.0 core of Codex CLI that must not be blurred during evaluation. Installation moved from npm to native installers (claude.ai/install.sh, install.ps1, brew cask, winget) and npm install -g @anthropic-ai/claude-code is officially deprecated, usually a way to escape the Node version matrix and global-permission support burden. The plugins/ directory is not a shell: it ships 13 officially maintained capability packs - code-review, pr-review-toolkit, commit-commands, feature-dev, frontend-design, security-guidance, agent-sdk-dev, plugin-dev, hookify, the explanatory and learning output styles, the opus 4.5 migration and ralph-wiggum - each a bundle of custom commands plus agents, turning personal prompts into installable team assets. It is also the reference harness for Anthropic models in third-party benchmarks (the Cursor footnote routes Anthropic model scores through Claude Code). Boundaries: no open license, closed models with unavoidable data egress, a terminal-first form factor that costs non-CLI users a learning curve, and a ceiling that depends on how well any third-party model adapts to the Anthropic protocol and tool calling when swapped in. The card metric is the verifiable fact (stars). Confidence: confirmed.

CONFIDENCE: Confirmed; Two or more independent sources, or reproduced through a harness
KEY METRIC: GitHub stars (repo has no LICENSE); Confirmed · 2026-09
MATURITY: Product; research → demo → product → production

Editor's take
Graded A (confirmed), and grade A here covers the factual layer only: repository stars and forks, the absence of a LICENSE, the four install channels and the npm deprecation, the names of the thirteen first-party plugins, the data-collection and "feedback is not used for training" clauses in the README, and Cursor's footnote naming it the comparator harness for Anthropic models. All of these come from checkable primary sources (repo README, docs tree, GitHub API, a third-party evaluation footnote). The entry asserts no SOTA capability reading for it, so there is no claim that would force a downgrade to C.

Water level: Claude Code's value has two layers. The first is that it became the industry's default shell for terminal agents - third-party evaluations use it as the execution environment for Anthropic models, while third-party model vendors (Kimi, Z.ai) use it as a host for theirs. Both holding true at once means its tool surface and protocol have leaked out into infrastructure. The second is that its plugin system productises engineering process itself: code review, a PR toolkit, commit commands, security guidance and model migration each ship as a pack, so teams distribute packages rather than folklore prompts.

Two things stated plainly: the repo is public but not openly licensed, which is not the same kind of reusability as Apache-2.0 Codex CLI; and the models are closed with data leaving the perimeter, a hard boundary for compliance-sensitive teams.

AI CodingCode HarnessTerminal AgentPlugins

AnthropicOfficial site github.com/anthropics/claude-code4 min read947

What it is: an agentic coding tool that lives in the terminal, and Anthropic's reference harness

Claude Code's own definition is deliberately modest: an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code and handling git workflows through natural language. The official docs name three surfaces: terminal, IDE, and tagging @claude on GitHub.

anthropics/claude-code stood at 148,438 stars / 24,902 forks when fetched on 2026-09-28, among the highest for tools in this class. One thing must be stated plainly: the repository has no LICENSE file (the GitHub API returns null), so it is publicly readable but not openly licensed - a material difference from Codex CLI's Apache-2.0 that should not be blurred during evaluation.

Install: native installers first, npm officially deprecated

The recommended install path has shifted to native installers: curl -fsSL https://claude.ai/install.sh | bash on macOS/Linux, irm https://claude.ai/install.ps1 | iex on Windows, with package managers via brew install --cask claude-code and winget install Anthropic.ClaudeCode. npm install -g @anthropic-ai/claude-code is explicitly marked deprecated. Then run claude in the project directory. A product that actively deprecates its npm channel is usually escaping the support burden of Node version matrices and global install permissions.

Plugins: thirteen first-party engineering packs shipped in the repo

The plugins/ directory is not a stub. It is a maintained set of capability packs, each combining custom commands with agents: code-review, pr-review-toolkit, commit-commands, feature-dev, frontend-design, security-guidance, agent-sdk-dev, plugin-dev, hookify, explanatory-output-style, learning-output-style, claude-opus-4-5-migration, ralph-wiggum.

The list is itself a statement about the product's shape. Code review, a PR review toolkit, commit commands, feature development, frontend design and security guidance are each their own pack. hookify implies hooks are a first-class extension point that tooling can generate. agent-sdk-dev and plugin-dev are extensions for people who build extensions. Two output styles (explanatory, learning) mean even response register is a swappable part. And claude-opus-4-5-migration turns "migrate to a new model" into a plugin - a clear signal of engineering process being productised.

Why it is the standard shell for Anthropic models in evaluations

Cursor's official Composer 2 evaluation footnote is explicit: on Terminal-Bench 2.0, Anthropic model scores use the Claude Code harness, OpenAI models use the Simple Codex harness, and Cursor itself runs the designated Harbor framework averaged over 5 iterations per model-agent pair. When a third party wants to measure the agent ceiling of a Claude-family model, Claude Code is the default shell.

Paired with Claude Opus 5.5 (released 2026-09-22, top of the Artificial Analysis intelligence index at 57.62, already catalogued here), that harness is the actual execution environment behind the reading. Coding agent performance is the product of model and harness, not the model alone - the precondition for reading any leaderboard in this space.

It is also a host for third-party models

Claude Code speaks the Anthropic Messages protocol, which by 2026 has become one of the de facto standards. Kimi's official documentation ships a dedicated guide for using Kimi models inside Claude Code and publishes Anthropic-compatible base URLs (https://api.kimi.com/coding/ in China, https://api.kimi.ai/coding/ overseas); Z.ai's coding plans likewise list Claude Code among the primary shells. The implication: part of Claude Code's moat is not the model but the fact that its tool surface and protocol are widely accepted - users can run someone else's model in its shell.

Data and privacy: the policy is written into the README

The README states collection scope directly: usage data (such as code acceptances or rejections), associated conversation data, and feedback submitted via /bug. It then lists three safeguards - limited retention periods for sensitive information, restricted access to user session data, and an explicit policy against using feedback for model training. Putting "we do not train on your feedback" in the repository README rather than burying it in a privacy policy gives enterprise procurement a clause it can quote. Feedback flows through the in-product /bug command plus an official Discord.

Boundaries

The repository is public but not openly licensed; models are closed and data leaves the perimeter; a terminal-first shape carries a learning cost for non-CLI users; and when a third-party model is substituted for Anthropic's, the ceiling depends on how well that model adapts to the Anthropic protocol and its tool-calling conventions.

More in AI Coding

SOTA hub

AI CodingDevelopmentTopC

ZCode: the official GLM-5.3 harness, defined as an ADE rather than an editor

ZCode is the official harness Z.ai built for GLM-5.3, and it defines itself as neither an AI editor nor a CLI but an ADE (Agentic Development Environment): it turns the 1M context window and long-horizon capability of GLM-5.3 into a stable desktop experience covering planning, coding, review and iteration, keeping goal, files, terminal output, browser context, execution mode and Git state inside one task so continuity survives from plan through implementation to verification, with model capability, tool calling and the execution chain tuned over multiple rounds against GLM-5.3. The most distinctive design is /goal mode: after a goal is set, every round ends with an automatic check of whether the goal is met, and the agent continues into the next round on its own until completion is confirmed (the documented example is a research task running twenty-plus rounds, with a right-hand panel showing what each round did). One session holds one goal at a time; /goal shows it, /goal sets it (replacing any existing goal), /goal replace swaps it explicitly, /goal pause suspends it, and the docs are honest about the fit - work that is easy to state in one sentence but takes many rounds to finish. The card metric keeps the verifiable price fact layer: GLM-5.3 API at $1.4 input / $4.4 output per million tokens. Boundaries: the client is not open source and the deep tuning only covers the GLM-5.3 family; a stable 1M context is a vendor claim and mid-context recall is an industry-wide weakness, so stuffing a whole repo is no substitute for retrieval; idle-time tasks are being rolled out to subscribers rather than always available; the free quota is a 5-day window, not a lasting benefit. Confidence C (vendor-claim).

$1.4 / $4.4GLM-5.3 API price (input / output, per million tokens)Vendor Claim · 2026-09

ProductZ.aiSite

LLMDevelopmentTopC

GLM-5.3: the open-weights coding flagship bought with post-training alone

Z.ai post-trained the very same base as GLM-5.2 and nothing else, producing the 744B-A40B GLM-5.3: 50 percent over 5.2 on the in-house Z.ai Code Bench, open-weights first on Terminal Bench 3.0 at 28.3 and Agents' Last Exam at 28.5, while AutomationBench 48.2 and GDPVal-AA v2 1769 beat every closed-source comparison on the official chart; CyberGym vulnerability discovery is SOTA and the exploitation-chain gains emerged without dedicated training. The companion 320B-A18B Flash starts from a new base and is the first GLM to mix sparse with linear attention, attacking long-context serving cost. Weights ship in FP8 and BF16, deployable on SGLang, vLLM, KTransformers and the Ascend stack.

28.3Terminal Bench 3.0 (top open-weight, vendor chart)Vendor Claim · 2026-08

ProductZ.ai (Zhipu AI)Site Repo

ReasoningDevelopmentTopA

GPT-6 Astra: first on terminal engineering tasks, third tier on composite intelligence

The OpenAI flagship tier, officially positioned for hardest end-to-end work (gpt-6-astra): 1.05M-token context, 128K-token single output, knowledge cutoff 2026-04-30, with Functions, Web search, File search and Computer use built in, so search, retrieval and interface control ship with the model instead of an outer agent framework. Reasoning effort runs low to max in five steps while price stays fixed at $10/$50, so the cost lever is token consumption; siblings Sol ($2/$10) and Luna ($0.1/$0.5) make a 100x spread, so load splitting can stay inside one vendor. Readings (2026-09-22): #1 on Terminal-Bench 4.0 at 58.18% (n_trials=330, pass@5 0.7121), the protocol closest to a coding agent's daily life; #2 on Arena Agent with net_improvement 11.54 and confirmed_success 17.70, both below leader Fable 5.1, but the field's highest praise at 32.79; #6 and #7 on the intelligence index (max 52.67). The cutoff is nearly five months older than this page, so version numbers and API changes must go through Web search; computer-use reliability appears on no board; closed, API only. Graded A (confirmed); not independently benchmarked.

58.18%Terminal-Bench accuracyConfirmed · 2026-09

ProductOpenAISite

LLMDevelopmentTopA

MiMo-V2.6: Live RL as an auditable ledger of 30 steps, 750k trajectories, and dollars

Xiaomi's open-source MiMo-V2.6 Pro/Flash: 30 steps of Live RL, roughly 750k trajectories, public cost of $850k/$2.62M, and 7k+ RL environments released in the same batch; tops the open-weights camp on the Artificial Analysis intelligence index at 46 with +17/+14 points out-of-sample on DeepSWE v1.1; four demo lines (Vibe World, CUA, research, content creation) all open. Self-improvement becomes an auditable ledger - steps, trajectories, and dollars on the table for third-party recomputation.

46Artificial Analysis Intelligence Index (top open-weight)Confirmed · 2026-09

ProductXiaomiSite Repo

As an Amazon Associate, we earn from qualifying purchases.