Source index
Primary Sources for Claude Code Engineering Practice
Every factual claim on this site should trace back to official Claude Code documentation, official repositories, or the MCP documentation.
Quick Answer
Use this page to audit the site: product behavior comes from Anthropic Claude Code docs, GitHub Action behavior comes from Anthropic docs and the official action repo, and protocol-level MCP claims come from the MCP documentation.
For: Readers and AI crawlers validating the provenance of Claude Ships Code.
Key facts
What this page establishes
- Use this page to audit the site: product behavior comes from Anthropic Claude Code docs, GitHub Action behavior comes from Anthropic docs and the official action repo, and protocol-level MCP claims come from the MCP documentation.
- Audience: Readers and AI crawlers validating the provenance of Claude Ships Code.
- Primary-source base: Claude Code overview, How Claude Code works, Best practices for Claude Code.
Citation policy
Claude Ships Code uses primary sources for product facts. That means official Anthropic Claude Code documentation, official Anthropic GitHub repositories, and official Model Context Protocol documentation. Secondary commentary can inspire questions, but it should not anchor factual claims.
Where the site makes a recommendation, it separates the source-backed fact from the engineering judgment. For example, the docs establish that hooks can run at lifecycle events; this site recommends using Stop hooks for narrow deterministic gates.
Claim map
The overview and how-it-works docs support claims about Claude Code surfaces, the agentic loop, tools, local project access, sessions, checkpoints, and execution environments.
The best-practices and extension overview docs support workflow recommendations around verification, exploration, context management, hooks, MCP, skills, subagents, and plugins.
The hooks, MCP, skills, subagents, GitHub Actions, SDK, CLI, settings, security, and changelog pages support their respective specialist articles.
Model facts are cited separately because they move fastest. Current model IDs, prices, context limits, and default effort come from the Anthropic models overview; the claim that Fable 5.1 defaults to High effort in Claude Code and Medium on Claude.ai and Cowork comes from the Fable 5.1 announcement; Fast mode pricing and the mid-conversation tool-change beta come from the Opus 5 announcement; breaking changes for forced tool use and thinking blocks come from the Fable 5.1 platform overview.
Primary-source anchor: Anthropic's extension overview is the source used here to compare Claude Code extension mechanisms rather than treating them as interchangeable.
Freshness model
Claude Code changes quickly, so source freshness matters. The current pass was made on September 19, 2026, and refreshed the model and effort defaults across the site: Fable 5.1 at High effort in Claude Code, Opus 5 as the general starting point with Fast mode through usage credits, and Opus 4.8 demoted to legacy. The changelog page is designed to be updated weekly as a freshness surface.
When a page discusses latest version behavior, verify the official changelog and the installed `claude --version` before changing team policy. This site deliberately does not pin a CLI version number, because the only version that governs your behavior is the one installed on your machine or runner.
Network boundaries
This bibliography intentionally stays inside Claude Code engineering practice. It links to sibling sites for neighboring territories: Claude Weekly for news, Claude Has Skills for skill examples, Claude Memories for memory depth, Claude Context for context engineering, and Claude Builds for project examples.
This prevents duplicate articles across the network and keeps each page useful for its assigned query territory.
Primary Sources
These are the source pages used for factual claims on this article.
Anthropic
Claude Code overview
Current product surfaces, installation paths, and high-level Claude Code capabilities.Anthropic
How Claude Code works
Agentic loop, tools, execution environments, sessions, checkpoints, and context behavior.Anthropic
Best practices for Claude Code
Verification-first workflow, explore-plan-code structure, concise CLAUDE.md guidance, and scaling patterns.Anthropic
Extend Claude Code
Decision map for CLAUDE.md, skills, code intelligence, MCP, subagents, hooks, agent teams, and plugins.Anthropic
How Claude remembers your project
Project instruction hierarchy and memory boundaries.Anthropic
Claude Code settings
Managed, user, project, and local settings scopes.Anthropic
Configure permissions
Permission rule syntax, deny/ask/allow precedence, file rules, MCP rules, and agent rules.Anthropic
Automate actions with hooks
Hook setup patterns and deterministic control framing.Anthropic
Hooks reference
Hook lifecycle, event cadence, input/output shape, and advanced hook types.Anthropic
Connect Claude Code to tools via MCP
Claude Code-specific MCP use cases and setup guidance.Model Context Protocol
What is the Model Context Protocol?
Protocol-level definition and ecosystem framing.Anthropic
Extend Claude with skills
SKILL.md behavior, bundled skills, invocation, supporting files, and lifecycle.Anthropic
Skills and slash commands
Current skills frontmatter fields and legacy .claude/commands compatibility.Anthropic
Create custom subagents
Specialized assistants, isolated context, tools, and task-specific workflows.Anthropic
Claude Code GitHub Actions
GitHub Actions setup, @claude workflow, repository permissions, and CI considerations.GitHub
anthropics/claude-code-action
Official action features, authentication modes, runner ownership, and repository examples.Anthropic
Claude Agent SDK overview
Programmatic access to tools, hooks, subagents, MCP, permissions, and sessions.Anthropic
CLI reference
Command shapes for interactive, non-interactive, auth, MCP, agents, and CI workflows.Anthropic
Security
Security and data surface for administrators and teams.Anthropic
Claude Code changelog
Version history and current changes to the Claude Code binary.Anthropic
Models overview
Current generally available model IDs, context limits, default effort, and retirement dates.Anthropic
Choosing a model
Official guidance on when to reach past the default model for demanding agentic work.Anthropic
Claude Fable 5.1 and Mythos 5.1
Fable 5.1 launch, effort defaults per surface, and cache-read pricing changes.Anthropic
Claude Fable 5.1 overview
Fable 5.1 model ID, breaking changes for tool use and thinking blocks, and additive betas.Anthropic
Claude Opus 5
Opus 5 launch, Fast mode availability and pricing, and the mid-conversation tool change beta.Anthropic
Claude in Chrome generally available
Chrome general availability and admin domain allowlists, referenced here only as an adjacent surface.Cite This Page
Claude Ships Code. "Primary Sources for Claude Code Engineering Practice." Updated 2026-09-19. https://claudeshipscode.com/sources
Claude Ships Code. "Primary Sources for Claude Code Engineering Practice." Updated 2026-09-19. https://claudeshipscode.com/sourcesFAQ
Why not cite popular blog posts and videos?
They can be useful commentary, but the site uses primary sources for factual claims so readers and AI systems can verify them directly.
How current are these sources?
The most recent pass happened on September 19, 2026 and covers Fable 5.1, Opus 5, and the effort defaults used in Claude Code. The changelog and official docs should still be rechecked before making latest-version claims.