Anthropic Claude Code vs Goose (2026)

Recipes are shareable packaged workflows with a documented YAML or JSON schema, sub-recipes compose them into larger workflows, subagents run them in isolated contexts and concurrently, and a scheduler runs them unattended.

The GitHub Action runs on any GitHub event or cron schedule without a mention, beside @claude mentions and terminal, desktop, IDE, web and headless runs.

A scheduler runs recipes on cron or a set frequency without a person starting them, on top of the desktop app, CLI, REST API and a gateway for remote access.

Grounding comes from direct file and command access and from MCP extensions that reach outside content, with project guidance read from AGENTS.md, and no codebase indexing, embedding or retrieval layer is documented.

Auto memory, written by Claude as it works and scoped per repository, persists across sessions until edited or deleted and is browsable and editable through /memory.

The built in Memory extension keeps what the user teaches goose as files per project or globally and loads them into every session, but the agent retains nothing on its own.

Manual Approval asks before any tool or extension use, Smart Approval flags riskier actions for approval and Chat Only allows no tool use, though Completely Autonomous is the default.

The agent runs locally under the user's own credentials with tools that can be disabled and scoped per recipe and subagent, and with no hosted service, accounts or organization layer it offers no SSO, RBAC, audit logging or attestation.

OpenTelemetry exports metrics, events and traces with a decision event for every permission prompt and a result event for every tool call, and cloud sessions log all operations for audit.

Every session is kept in a local SQLite database with the full conversation of messages and tool interactions, browsable and resumable in the CLI and desktop app, though with no organization level export.

Claude Code runs on the developer's own machine by default, cloud sessions run in isolated Anthropic VMs or on the organization's self hosted environments, and data is encrypted in transit but not at rest.

Plugins bundle skills, subagents, hooks and MCP servers and are browsable in Anthropic's official marketplace, beside bundled skills and built in subagents.

A published Recipe Cookbook offers ready to use recipes shareable by URL and launchable in one click, alongside more than 70 installable MCP extensions and a skills system for role based specialization.

Every model is Anthropic's: customers choose a Claude model and where it is served, but not the model maker.

claude plugin eval runs prompt suites in fresh isolated sessions, grades them against a no plugin baseline and can gate CI on the score.

The repository's evals/harbor directory runs goose against benchmark tasks so a customer can evaluate the agent's behavior with the project's own harness, and recipes add structured response schemas and retries for deterministic checks.

The Chrome integration lets the agent operate a visible browser session with the user's logins, and computer use extends this to native macOS apps.

Computer control comes from the built in Computer Controller extension, which drives macOS apps through the third party Peekaboo CLI and scrapes the web, while browser control comes from catalog extensions such as Playwright, Selenium and Browserbase that the user installs.