Today Explore Practice Library Exam Certificate

Noesis

Deep, finishable courses on prompting, psychology, money, health, and more — hands-on, in English and Khmer.

Learn

Today Explore Practice Library All courses

Certify

Take the exam Certificate

Account

My account Sign in

© 2026 Noesis. All rights reserved.Built for people who never stop learning.

Today Explore Practice Library Profile

Noesis — Deep, finishable courses.

Claude Code

It doesn't just answer.
It runs the loop.

25 lessons on the agentic loop, context budgeting, permissions, skills, hooks, MCP, subagents, and unattended automation.

💭 Honest take

24 lessons across 6 tracks, every one built on the full learning loop: a real failure, a mental model, an expert trace, an attempt-first checkpoint, a guided lab, and a close that names the decision rule. Content is grounded in the official Claude Code documentation and each lesson links its sources. Claude Code ships quickly, so treat specific flags, settings keys, and version notes as things to verify against the current docs — the reasoning behind them changes far more slowly. Lesson bodies are authored in English; Khmer localization still needs fluent review.

Start learning Explore topics

25

lessons

0

real prompts

3

models covered

6

role perspectives

Current track 1. Core Workflow0 / 5 · ~20 min left

Overall Progress: 0 / 25 lessons completed

✓Close the agentic loop — give every task a check the agent can run instead of stopping at "looks done"

✓Treat the context window as a budget: measure it, clear it, compact it, and keep CLAUDE.md short enough to be followed

✓Write permission rules that actually hold, and bound the blast radius of untrusted input and prompt injection

✓Route each need to the right mechanism — CLAUDE.md, skills, hooks, MCP, or a plugin — instead of the familiar one

✓Delegate to subagents, run an adversarial review, and work in parallel without sessions colliding

✓Run unattended work safely: bounded non-interactive runs, batch fan-out, scheduled jobs, and a team rollout you can audit

Up next

Your First Verified Fix

Make one tiny change and watch a test prove that it works — Beginner

🎯 Learning Tracks

Curated paths through the course — pick one and finish it, instead of scrolling a thousand lessons.

0 / 6 complete

Start here01Core Workflow0/5 lessons·Start 02Context and Project Instructions0/4 lessons·Start 03Permissions and Safety0/4 lessons·Start 04Optional: Extensions0/4 lessons·Start 05Optional: Delegation0/4 lessons·Start 06Optional: Automation0/4 lessons·Start

Beginner

0 / 7

Your First Verified Fix

Make one tiny change and watch a test prove that it works

What Fills the Context Window

Almost every best practice in this course exists because of one constraint

The Agentic Loop

See the read → act → check loop behind the fix you just made

Clear, Compact, Rewind

Three ways to reclaim context, and the one that loses the thing you needed

Explore, Plan, Code, Commit

Separating research from execution stops you from perfectly solving the wrong problem

Prompting for Precision

Scope, sources, patterns, and evidence — four moves that remove most corrections

Course-Correct and Recover

Esc, rewind, and clear — and the rule that says which one you need

Intermediate

0 / 6

Choose the Right Extension

Six mechanisms, six triggers — most setups fail by picking the familiar one

The Permission Model

Enforced by Claude Code, not by the model — and that distinction is the whole safety story

Skills That Load When Needed

A SKILL.md is a folder, a description, and a body — the description does most of the work

CLAUDE.md That Gets Followed

A bloated instructions file is worse than no instructions file

Auto Memory and Scoped Rules

Notes Claude writes for itself, and instructions that only load when they matter

MCP Without the Sprawl

Connect the systems Claude cannot reach — and only those

Advanced

0 / 12

Subagents and Context Isolation

Send the reading somewhere else and get back only what you need

Non-Interactive Mode

claude -p turns the agent into a Unix command — with all that implies

Designing a Custom Subagent

A markdown file with frontmatter that turns a repeated delegation into a role

Fan-Out and Batch Work

Two thousand files, one prompt, and the three files you test it on first

Writing Rules That Hold

Deny beats ask beats allow — and argument-matching patterns are weaker than they look

Adversarial Review

A reviewer that never saw the reasoning judges the result on its own terms

Hooks and Guarantees

The only layer that runs whether the model agrees or not

Scheduled and CI Work

Cloud routines, local scheduled tasks, CI actions, and in-session loops

Modes and Sandboxing

Choose how often you are asked — and what happens when nobody is there to answer

Parallel Sessions and Worktrees

More than one Claude at once, without two of them editing the same file

Team Rollout and Governance

What you commit, what you enforce centrally, and what stays personal

Untrusted Content and Injection

Everything Claude reads is data — including the parts that look like instructions