Agentic Coding in a Minute

Learning paths

Follow a sequence

Start with a curated path, then branch into related lessons as you need them.

• Claude Code FundamentalsFrom first session through project instructions, permissions, planning, Git, testing, and MCP.8 lessons
• Codex FundamentalsCodex-oriented setup: instructions, tasks, sandboxing, Git, verification, MCP, and automation habits.8 lessons
• Git for Coding AgentsThe Git skills that matter when agents commit, branch, recover, and work in parallel.9 lessons
• Agentic Engineering FundamentalsVendor-neutral loop: specify, plan, context, tools, execute, verify, retry, and approve.9 lessons

All learning paths →

Tracks

Browse by topic

Expandable tracks covering tools, techniques, and engineering practices for agentic development.

• Claude Code in a MinutePractical Claude Code usage from project instructions through advanced agent workflows.6 lessons
• Codex in a MinuteCodex CLI and related agentic coding workflows, starting with project instructions and task setup.5 lessons
• Git in a MinuteGit concepts that matter most when coding agents commit, branch, and work in parallel.9 lessons
• MCP in a MinuteModel Context Protocol: servers, tools, permissions, and how coding agents use them.4 lessons
• Context Engineering in a MinuteHow to feed agents the right context—and keep it from rotting mid-task.3 lessons
• Agent Workflow in a MinuteVendor-neutral engineering patterns for planning, verification, and multi-agent work.7 lessons
• Testing with Agents in a MinuteVerification gates, agent-written tests, and how to avoid fake green builds.3 lessons

All lessons

Start anywhere

• Claude Code · beginnerGetting startedClaude Code is Anthropic’s agentic coding CLI: you give a task in a repo, it reads context, edits files, runs commands, and iterates with your approval boundaries.
• Claude Code · beginnerCLAUDE.mdCLAUDE.md is a project instruction file Claude Code reads so it follows your repo’s conventions without being reminded every session.
• Claude Code · beginnerPlanningAsk Claude Code for a short plan and file list before it edits, so you can reject bad approaches cheaply.
• Claude Code · intermediatePermissionsClaude Code permissions define which tools and actions the agent may take without asking—and which require human approval.
• Claude Code · intermediateGit workflowUse a branch-per-task Git workflow with Claude Code so agent commits stay reviewable and reversible.
• Claude Code · advancedHooksClaude Code hooks are automation points around agent events—use them for checks and guardrails, not mystery side effects.
• Codex · beginnerGetting startedCodex is OpenAI’s coding agent surface for repository work—task in, inspected changes out, with project instructions guiding behavior.
• Codex · beginnerAGENTS.mdAGENTS.md is a repository instruction file for coding agents (including Codex) that records how to work in the project.
• Codex · beginnerTask specificationA good Codex task states outcome, constraints, and verification—so the agent can finish without inventing scope.
• Codex · intermediateSandboxingSandboxing limits what Codex can touch—filesystem, network, and privileges—so agent speed does not equal unrestricted access.
• Codex · advancedAutomationAutomate Codex only after the human loop works: same task template, same gates, same review boundary.
• Git · beginnerCommitsA commit snapshots a set of changes with a message—your unit of history for review and rollback.
• Git · beginnerBranchesA branch is a movable pointer to a commit—your named line of history for a change.
• Git · beginnerDiffsA diff shows what changed between snapshots—your primary review tool for agent output.
• Git · intermediateRebaseRebase replays commits onto a new base—useful for updating a branch, dangerous on shared history.
• Git · intermediateWorktreesGit worktrees let you check out multiple branches of the same repository into separate directories at once.
• Git · intermediateCherry-pickCherry-pick copies a commit onto your current branch—useful for rescuing a good agent fix from a messy branch.
• Git · advancedParallel agent workspacesRun multiple coding agents safely by giving each a worktree, branch, and clear merge path.
• Git · intermediateBisectgit bisect binary-searches history to find which commit introduced a bug—vital when agent commits are frequent.
• Git · intermediateRecoveryGit recovery tools—reflog, reset, revert—help undo agent mistakes without panicking.
• MCP · beginnerWhat MCP isMCP (Model Context Protocol) is a standard way for AI clients to connect to external tools and data through servers.
• MCP · intermediateTools and resourcesIn MCP, tools are actions the model can invoke; resources are data the model can read.
• MCP · intermediateLocal vs remote serversLocal MCP servers run near your machine; remote servers expose tools over the network—trust and latency differ.
• MCP · intermediatePermissions and securityMCP security is mostly about which servers you trust, which tools they expose, and what those tools can reach.
• Context · beginnerProject instructionsProject instructions are durable repo-level guidance that coding agents load so they do not rediscover your process every session.
• Context · intermediateRelevant file selectionRelevant file selection means loading only the files needed for the task—so the agent reasons over signal, not the whole repo.
• Context · intermediateContext pollutionContext pollution is when irrelevant files, logs, or tool output crowd the model’s window and degrade the next decision.
• Workflow · beginnerTask specificationA task specification defines outcome, scope, constraints, and verification for any coding agent—not just one vendor.
• Workflow · beginnerPlan before executionHave the agent propose a short plan and file touch-list before it edits—so you can catch wrong approaches cheaply.
• Workflow · intermediateHuman approvalHuman approval boundaries are checkpoints where an agent must stop for a person before risky or irreversible actions.
• Workflow · intermediateExecutionExecution is the agent actually editing and running commands—bounded by the plan, permissions, and verification gate.
• Workflow · intermediateVerification loopsA verification loop requires the agent to prove done with an external check—then fix—until the check passes or you stop it.
• Workflow · intermediateRetriesRetries are bounded re-attempts after a failed verification—each attempt should use new evidence, not blind repetition.
• Workflow · advancedParallel agentsParallel agents only help when tasks are partitioned, isolated on disk, and integrated through reviewable merges.
• Testing · beginnerVerification gatesA verification gate is a deterministic check that must pass before agent work is accepted.
• Testing · intermediateTest-first agent workflowsTest-first agent workflows make the agent write or confirm a failing check before implementing—so done means the check flipped green.
• Testing · intermediateAvoiding self-verificationDo not let an agent be the only judge of its own correctness—especially when it can edit the tests.