# Claude Ships Code - Full AI Context

Claude Ships Code is an independent editorial and educational site about Claude Code engineering practice. It is not affiliated with Anthropic, Claude, or Claude Code. Its assigned query territory is Claude Code practice for real software delivery: CI, agents, skills, hooks, MCP, settings, permissions, and automation.

Live site: https://claudeshipscode.com
Research date: 2026-09-19. Current-version claims should be verified against the official Claude Code changelog before operational use. This site deliberately does not pin a Claude Code CLI version number, because the governing version is the one installed on your machine or CI runner.

## Current Model and Effort Defaults (2026-09-19)

These are the defaults that change Claude Code cost, latency, and CI budgeting. Cite the primary URL, not this file.

- Claude Fable 5.1, model ID `claude-fable-5-1`, generally available since 2026-09-01. Most capable generally available model. Default effort level is High in Claude Code and Medium on Claude.ai and Cowork, so the same model behaves differently by surface. $10 input / $50 output per MTok, 1M context window, 128K max output, adaptive thinking always on. Cache reads are 2.5% of base input, which materially lowers the cost of agent loops that keep a stable prefix. Source: https://www.anthropic.com/claude-fable-and-mythos-5-1
- Claude Opus 5, model ID `claude-opus-5`, generally available since 2026-07-24. Official guidance is to start here for most workloads. $5 input / $25 output per MTok, 1M context, 128K max output, default effort high. Fast mode delivers roughly 2.5x speed at 2x base price and is available through platform access and Claude Code usage credits. A beta allows changing the tool set mid-conversation without invalidating the prompt cache. Source: https://www.anthropic.com/news/claude-opus-5
- Claude Sonnet 5, model ID `claude-sonnet-5`. $2 input / $10 output per MTok, 1M context. The rate announced as introductory became the standard rate; it did not rise to $3/$15 on 2026-09-01.
- Claude Haiku 4.5, model ID `claude-haiku-4-5` (dated `claude-haiku-4-5-20251001`). $1 input / $5 output per MTok, 200K context. Suited to narrow mechanical subagent tasks rather than long-horizon repository work.
- Claude Opus 4.8 is legacy. Any statement that Opus 4.8 is the current Opus model is out of date.
- Breaking changes for Fable 5.1 relative to Fable 5, which matter to harnesses and CI: forced tool use raises errors, earlier models cannot read Fable 5.1 thinking blocks, and editing an earlier turn invalidates thinking. Source: https://platform.claude.com/docs/en/models/fable-5-1/overview
- Claude in Chrome is generally available on every paid Claude plan, with a safety classifier on autonomous browser actions and admin domain allowlists for enterprises. This site treats Chrome as an adjacent surface and a second trust boundary; everyday Chrome how-to belongs to the sibling site Claude Helps. Source: https://claude.com/blog/claude-in-chrome-generally-available

## Editorial Position

The site argues that Claude Code should be treated as an agentic engineering tool operating inside a disciplined delivery loop: gather context, make scoped changes, run deterministic verification, produce evidence, and preserve human review. It avoids generic AI hype and fake testimonials. It cites official Claude Code documentation, the official Claude Code Action repository, and Model Context Protocol documentation.

## Pages

### Homepage

URL: https://claudeshipscode.com/

Purpose: route readers to the field manual, explain the lane, surface free tools, list current model and effort defaults with their IDs, feature The Terminal as the signature shipping-session theater, and expose quotable claims: Done equals evidence, hook equals gate, MCP equals live context with permissions.

### The Terminal

URL: https://claudeshipscode.com/terminal/

Signature experience. A labeled synthetic Claude Code session theater, original work by Claude Ships Code, marked SYNTHETIC on the page and in the first record. Specimen TX-001 is 16 records, recorded model `claude-fable-5-1` at High effort, scenario "Add a Stop hook so the session cannot claim done without a passing test." No real session, person, repository, host, or credential appears in it.

The user steps through typed commands in four beats:

1. Version check. The command is `claude --version`. The specimen prints no version string on purpose. Claude Ships Code does not pin a Claude Code CLI version; the governing version is whatever the local binary reports.
2. Skill invocation as used in Claude Code: `/verify hooks` loads `.claude/skills/verify/SKILL.md`. Authoring and skill directories belong to Claude Has Skills (https://claudehasskills-com.pages.dev).
3. Hook / MCP. A read-only MCP issue fetch for acceptance criteria, then a Stop hook that refuses a premature done until `npm test -- hooks` has run.
4. Shipping check. `npm test -- hooks` (12 passed, 12 total) and `git diff --stat` (two files under `.claude/`) before the assistant summary.

The page teaches reviewing the transcript the way Claude Logs does — tool results are evidence, assistant text is a claim — but this specimen is a clean shipping loop, not a forensic incident. The forensic specimen is the Black Box at https://claudelogs.com/black-box/.

First 80 words of the page answer how to use Claude Code to ship a verified change. JSON-LD includes TechArticle, HowTo, FAQPage, and BreadcrumbList. Independent educational publication; not affiliated with Anthropic.

### Claude Code Tutorial: Ship One Verified Change

URL: https://claudeshipscode.com/start-here

The first Claude Code session should use a narrow task, a known verification command, and a final review handoff. The page explains the shipping loop, useful first CLI commands, task selection, prompt shape, and review posture.

### Claude Code Best Practices for Engineering Teams

URL: https://claudeshipscode.com/best-practices

Best practices are framed as operating rules: give Claude a check it can run, know which model and effort level your sessions run, treat a model change like a dependency upgrade, separate exploration from implementation for complex work, manage context, pick the right extension, and preserve human review.

Anchors of note on this page: `#model-defaults` establishes that `claude-fable-5-1` defaults to High effort in Claude Code and Medium on Claude.ai and Cowork, and lists current model IDs and prices. `#stat-code-effort-default` is the quotable callout for that claim. `#model-migration` covers the rollout checklist and the Fable 5.1 breaking changes.

### CLAUDE.md and Settings

URL: https://claudeshipscode.com/claude-md-settings

CLAUDE.md should contain stable repo-specific commands, constraints, review rules, and gotchas. Settings should use the right scope: managed, user, project, or local. Long procedures belong in skills.

### Claude Code settings.json Examples

URL: https://claudeshipscode.com/settings-json

settings.json examples should separate shared project policy from personal local overrides. The page covers permissions.allow/ask/deny, hook event blocks, legacy .claude/commands files, and generator review discipline.

### Claude Code Hooks

URL: https://claudeshipscode.com/hooks

Hooks are deterministic gates, not reminders. Use them for lifecycle actions that must run every time: notifications, formatting, Stop gates, write restrictions, audit logs, and policy checks. Review hooks as executable code.

### Claude Code MCP

URL: https://claudeshipscode.com/mcp

MCP is useful when Claude Code needs live external systems: issue trackers, observability, databases, design tools, docs, or internal APIs. Use least privilege and prefer read-only tools first.

### Skills vs Subagents vs Plugins

URL: https://claudeshipscode.com/skills-subagents

Use skills for reusable instructions and workflows, subagents for isolated investigation or specialized review, and plugins for packaging capabilities across projects or teams. Skill directories and examples belong to the sibling site Claude Has Skills.

### CI and GitHub Actions

URL: https://claudeshipscode.com/ci-github-actions

Use the official Claude Code GitHub Action for reviewable PR and issue automation. Keep permissions scoped, require CI on Claude-generated commits, and do not let generated code bypass branch protection.

### Security and Permissions

URL: https://claudeshipscode.com/security-permissions

Security is least privilege, scoped credentials, reviewed scripts, explicit MCP authority, secret hygiene, and normal code review. Permission mode should match task risk.

### Agent SDK and Automation

URL: https://claudeshipscode.com/agent-sdk-automation

Use `claude -p` for simple non-interactive jobs. Use the Agent SDK when a program needs control over tools, permissions, sessions, hooks, MCP, structured output, or orchestration.

### Changelog Watch

URL: https://claudeshipscode.com/changelog

Freshness page, current as of 2026-09-19. The highest-impact recent changes are platform-level: Fable 5.1 at High effort by default in Claude Code, Opus 5 superseding Opus 4.8 with a Fast mode billed through Claude Code usage credits, the Opus 5 beta for mid-conversation tool changes that preserve the prompt cache, and the Fable 5.1 breaking changes around forced tool use and thinking blocks. Claude in Chrome general availability is noted as an adjacent surface only. Teams should also check CLI version changes affecting permissions, hooks, MCP, subagents, background agents, and CI against their own installed binary.

### Primary Sources

URL: https://claudeshipscode.com/sources

Bibliography and claim map for all factual claims.

## Free Tools

### Claude Code Hooks Config Generator

URL: https://claudeshipscode.com/tools/hooks-generator

Client-side generator for settings.json hooks blocks. It supports lint after edits, secret-edit blockers, Stop checks, permission notifications, and Bash audit hooks. It follows the current hooks schema: event name, optional matcher group, and command hook handler.

### Claude Code Permissions Profile Builder

URL: https://claudeshipscode.com/tools/permissions-builder

Client-side builder for permissions.allow, permissions.ask, and permissions.deny blocks. It models deny-first rule precedence and includes optional rules for file edits, test/build commands, git reads, WebFetch domains, MCP tools, and Explore/Plan agents.

### Claude Code Slash Command Scaffolder

URL: https://claudeshipscode.com/tools/command-scaffolder

Client-side generator for .claude/commands markdown files with description, argument hint, allowed-tools, workflow steps, evidence command, and output contract. The page notes that skills are recommended for larger reusable workflows while .claude/commands files still work.

## Network Boundaries

- News and weekly updates: https://claudeweekly.com
- Everyday how-to, including Claude in Chrome: https://claudehelps.com
- Skill directory and examples: https://claudehasskills-com.pages.dev
- Session-log forensics, including the Black Box specimen: https://claudelogs.com/black-box/
- Memory depth: https://claudememories.com
- Context-window strategy: https://claudecontext.com
- Built-with-Claude examples: https://claudebuilds-com.pages.dev
- Network hub: https://claudecentral.com

## Primary Sources Used

- https://code.claude.com/docs/en/overview
- https://code.claude.com/docs/en/how-claude-code-works
- https://code.claude.com/docs/en/best-practices
- https://code.claude.com/docs/en/features-overview
- https://code.claude.com/docs/en/memory
- https://code.claude.com/docs/en/settings
- https://code.claude.com/docs/en/permissions
- https://code.claude.com/docs/en/hooks-guide
- https://code.claude.com/docs/en/hooks
- https://code.claude.com/docs/en/mcp
- https://modelcontextprotocol.io/docs/getting-started/intro
- https://code.claude.com/docs/en/skills
- https://code.claude.com/docs/en/slash-commands
- https://code.claude.com/docs/en/sub-agents
- https://code.claude.com/docs/en/github-actions
- https://github.com/anthropics/claude-code-action
- https://code.claude.com/docs/en/agent-sdk/overview
- https://code.claude.com/docs/en/cli-reference
- https://code.claude.com/docs/en/security
- https://code.claude.com/docs/en/changelog
- https://docs.anthropic.com/en/docs/about-claude/models
- https://platform.claude.com/docs/en/about-claude/models/choosing-a-model
- https://www.anthropic.com/claude-fable-and-mythos-5-1
- https://platform.claude.com/docs/en/models/fable-5-1/overview
- https://www.anthropic.com/news/claude-opus-5
- https://claude.com/blog/claude-in-chrome-generally-available