Claude Code

Context

Reads repository files and command output; can use CLAUDE.md-style project instructions, imported context, MCP resources, and resumed session history.

Actions

Edits files, runs shell commands, works with Git, calls configured MCP tools, and can be scripted through print mode and the SDK.

Execution

Primarily executes in the local environment where the CLI is launched; external tools and MCP servers have their own execution boundaries.

Memory & rules

Project and user instructions live in CLAUDE.md and related configuration. Session resume and optional memory features are product-managed context, not changes to the base model.

Permissions

Permission modes and allow/deny settings control tool use. Anthropic exposes a dangerous bypass flag and explicitly warns that it skips prompts.

Verification

Can run tests, linters, type checkers, and Git inspection commands, but the user must choose the checks and review the resulting diff.

Best for

• Command-heavy repository work
• Multi-file changes with an active human reviewer
• Teams already using Anthropic models and Claude Code instructions

Limitations

• Editorial judgment: visual work when no browser or image tool is connected
• Editorial judgment: unattended production actions without an external approval boundary