Claude Fundamentals

What it is

Term	Meaning
Agentic coding tool	Coding agent that reads your codebase, edits files, runs commands (Claude Code, Codex, Kiro, OpenCode)
Code harness	Program around an LLM that makes it an agent: prompts, tools, sandbox (Claude Code, Aider, Cursor, Cline…)
Claude Code	CLI agent; also in IDE, desktop, web. Site: code.claude.com

Agentic Loop

Prompt → Model → [Gather Context → Take Action → Verify] (loop) → Done

Each phase uses tool calls (code functions that call external programs).
Interrupt/steer any time (Esc).

Built-in Tools

Tool names are the exact strings used in permission rules, subagent tools lists and hook matchers. Needs permission: Bash, PowerShell, Edit, Write, NotebookEdit, ExitPlanMode, Skill, WebFetch, WebSearch.

Category	Tools
Files	Read, Edit, Write, NotebookEdit
Search	Glob, Grep (absent by default on macOS/Linux/WSL, where search runs through Bash)
Shell	Bash, PowerShell (native Windows without Git Bash), Monitor (background command, streams each output line back)
Web	WebFetch, WebSearch
Code intel	LSP (jump to definition, references, type errors after edits)
Agents	Agent, SendMessage, ListAgents, SubagentHandback (auto mode), Workflow (dynamic workflows)
Planning/worktrees	EnterPlanMode, ExitPlanMode, EnterWorktree, ExitWorktree
Tasks	TaskCreate / TaskGet / TaskList / TaskUpdate / TaskStop (model-dependent). `TodoWrite` is off by default; `TaskOutput` was removed in v2.1.277
Scheduling	CronCreate / CronList / CronDelete (session-scoped), ScheduleWakeup (`/loop`), RemoteTrigger (cloud scheduled tasks)
MCP	ListMcpResourcesTool, ReadMcpResourceTool, WaitForMcpServers, ToolSearch
Interaction	AskUserQuestion, PushNotification, SendUserFile, SendFeedback, ReportFindings
Artifacts	Artifact (+ ArtifactComments / ArtifactData)
Other	Skill, EndConversation (rare abuse case), ShareOnboardingGuide (`/team-onboarding`)

Rule formats: Bash(npm run *) also covers Monitor; Read(...) covers Read/Grep/Glob/LSP; Edit(...) covers Edit/Write/NotebookEdit.

Models

Alias	Resolves to
`default`	Account default (Opus 5.5 on most plans)
`best`	Same as `fable` where available, else `opus`
`fable`	Fable 5.1
`opus`	Opus 5.5
`sonnet`	Sonnet 5.5
`haiku`	Haiku 5.5
`sonnet[1m]`, `opus[1m]`	1M-context variants
`opusplan`	Opus in plan mode, Sonnet for execution (`opusplan[1m]` for 1M in both)

Pin a version with a full ID (claude-opus-5-5, claude-fable-5-1, claude-haiku-5-5) or ANTHROPIC_DEFAULT_OPUS_MODEL / _SONNET_ / _HAIKU_ / _FABLE_MODEL.
On Bedrock, Vertex and Foundry the aliases can resolve to older models (e.g. sonnet → Sonnet 4.5), so pin explicitly there.
Minimum Claude Code versions: Fable 5.1 v2.1.257, Opus 5.5 v2.1.280, Sonnet 5.5 v2.1.284, Haiku 5.5 v2.1.293.

Tiers: Fable = top capability, Opus = deep reasoning, Sonnet = balanced, Haiku = fast/cheap.

claude --model opus      # whole session
/model sonnet            # switch mid-session

{ "model": "opus", "availableModels": ["sonnet", "haiku"] }

Each model has its own prompt cache, so the first request after a switch re-reads the conversation uncached.

Effort

Set with /effort, --effort, CLAUDE_CODE_EFFORT_LEVEL, or effortLevel in settings (skills and subagents can set effort in frontmatter).

Model	Levels	Default
Fable 5.x	low, medium, high, xhigh, max	high
Opus 5.5 / Sonnet 5.5 / Haiku 5.5	low, medium, high, xhigh, max	medium
Opus 5, Sonnet 5, Opus 4.8	low … max	high
Opus 4.7	low … max	xhigh
Opus 4.6, Sonnet 4.6	low, medium, high, max	high

An unsupported level falls back to the highest supported one at or below it (xhigh → high on Opus 4.6).
max applies to the current session only unless set via the env var.
ultracode is a separate Claude Code setting (dynamic workflows), set with /effort ultracode, not an effort level.

Fast Mode

/fast — up to 2.5x faster Opus at a higher per-token price (research preview). Same model, different API configuration.

Supported: Opus 5.5 (default), Opus 5, Opus 4.8. Not Sonnet, Haiku or Opus 4.7.
Price per MTok in/out: $8/$40 on Opus 5.5, $10/$50 on Opus 5 / 4.8. Flat across the 1M window.
Subscriptions: billed from usage credits only, not plan limits (turn on with /usage-credits). Team/Enterprise: an Owner must enable it first.
Not on Bedrock, Vertex or Foundry.
Enable at the start of a session (switching mid-way re-reads context uncached). ↯ icon shows it's on.
"fastMode": true persists it; fastModePerSessionOptIn resets each session; CLAUDE_CODE_DISABLE_FAST_MODE=1 disables it.

Modes

Normal: runs, asks only for specific commands.
Plan: writes a plan to review → "Yes, auto-accept" / "Yes, manually approve edits" / "No, keep planning".
Shift+Tab cycles permission modes, which include Plan and Auto (see Claude Permissions).

Other

Images: drop screenshots/diagrams into chat to analyze or generate code (not draggable in WSL2 terminals).
Resources: free courses at academy.claude.com (Claude Code 101, AI Fluency and others; the older skilljar.com catalog may have moved).