╱╲
╱ ╲
╱ ◆◆ ╲
╱ ◆◆ ╲
╱ ◆◆ ╲
╱ ◆◆ ╲
╱─────◆◆─────╲
◆◆
◆◆
[running...]
A demonstration in two acts
Stop
writing
code.
Start
directing
it.
Claude Code is an agentic coding tool that lives in your terminal — reads your repo, edits real files, runs your tests, and tells you when it's wrong. This page is a showcase. It was also built by it.
[ ACT I // 01 ]
A real session, replayed
PRESS A SCENARIO BELOW
[ ACT I // 02 ]
What it actually does
SIX CAPABILITIES
01 // AGENTIC LOOP
It plans. It edits. It runs the tests. It fixes its own mistakes.
Give it a goal — not a recipe. Claude Code reads the relevant files, makes a plan, edits the code, runs your test suite, watches the output, and iterates until the bar is green. You stay in the loop; you don't have to drive every keystroke.
→ /plan, /run, /think harder
200k
CONTEXT WINDOW
Whole repo, in one head. No more rummaging tab to tab — Claude Code reads the files it needs and remembers what it found.
02 // TERMINAL-NATIVE
No IDE. No extension. Just your shell.
Lives in the terminal you already have. Pipes in. Pipes out. Composes with everything.
npm i -g @anthropic-ai/claude-code
03 // MCP-AWARE
Tools that belong to your stack.
Connect Postgres, GitHub, Sentry, your internal API — anything that speaks Model Context Protocol. Claude reaches for them like a senior engineer reaching for a runbook.
→ /mcp
04 // HOOKS & SUBAGENTS
Customize the loop itself.
Pre-tool hooks. Post-tool hooks. Spawn subagents that run in parallel and report back. Bend the agent to fit your team's idioms.
.claude/hooks/
05 // CLAUDE.md
A constitution for your codebase.
Drop a CLAUDE.md at the repo root and the agent reads it like a tenured engineer reads onboarding docs. Style guides. Architectural choices. The "we don't do that here" list. It listens.
/init
06 // CHECKPOINTS
Every diff is reversible.
Every change is checkpointed. Roll back a mistake with a single command. Claude Code is willing to be bold because you have an undo button as long as your patience.
/rewind
[ ACT II // 01 ]
A day with Claude Code
FOUR MOVES
i. MORNING
Review the overnight runbook.
Spawn a subagent on yesterday's failing pipeline. While you make coffee, it bisects, isolates, and writes the postmortem.
$ claude # pick up where you left off
ii. MIDDAY
Ship the feature.
Sketch intent in plain English. Watch it scaffold the route, the migration, the tests, and the docs. Push back where it's wrong. It listens.
> add /v2/incidents endpoint with severity routing
iii. AFTERNOON
Pair on the gnarly bug.
A 200k-token window means it reads the whole call graph. You think out loud. It thinks back. Sometimes it's right. Sometimes you are. Either way, faster.
> think harder about the race condition in the consumer
iv. EVENING
Hand off to the loop.
Set up an autonomous run. Refactor the auth module across nine services. Wake up, review the diff, merge.
$ claude --print "refactor JWT validation per CLAUDE.md"
The compiler used to translate our intent into instructions. Claude Code translates our intentions into outcomes. The interface to software has moved up the stack.
— a thought, while waiting for tests to pass