agenticlately · GH-600 Study Prep
Unofficial GH-600 exam prep

GH-600 Prep

Story → idea → example → quiz.
A self-study companion built one lesson at a time.

Self-paced · most learners finish in 2 to 4 weeks · free, no signup

Phase 0 · GitHub fluency

— / —

Classic GitHub. Prereq for every later phase. No agents yet.

Phase A · Prepare agent architecture and SDLC processes

— / — 15–20%

GitHub surface for D1 first (Issues + PRs as agent artifacts), then the agent layer.

Phase B · Implement tool use and environment interaction

— / — 20–25%

Highest weight on the exam. GitHub Actions deep dive, then MCP + Copilot CLI.

B0 lockedGitHub Actions deep diveEvents, contexts, expressions.
2.1 locked4-layer tool stackContext → filter → MCP → tools.
2.2 lockedMCP fundamentalsThe USB-C of AI tools.
2.3 lockedMCP in practiceGitHub MCP server live.
2.4 lockedAllow lists & registriesPermission boundary.
2.5 lockedRead-only vs implementationThe trap MCQ.
2.6 lockedCopilot CLI surfaceSlash commands + flags.
2.7 lockedActions for agent validationHow agents consume runs.
2.8 lockedCustom agents configVerbatim YAML keys.

Phase C · Manage memory, state, and execution

— / — 10–15%
C0 lockedWorkflow artifacts + needsClassic GitHub state handoff.
3.1 lockedWhat "state" means for agentsMemory vs session vs artifacts.
3.2 lockedCopilot memory modelWhat persists across runs.
3.3 lockedSession checkpointingResume across machines.
3.4 lockedArtifacts as durable stateHow agents persist work.
3.5 lockedExecution environmentsSandbox vs full repo vs branch.

Phase D · Perform evaluation, error analysis, and tuning

— / — 15–20%

Hooks live here. Beta-taker priority: learn types + when they fire.

D0 lockedCode + secret scanning basicsClassic GitHub safety nets.
4.1 lockedWhy evaluate agents differentlyNon-determinism, drift.
4.2 lockedImplementation Planner patternCanonical planning agent.
4.3 lockedError analysis methodologyLogs, retries, instrumentation.
4.4 lockedScanning as agent safety netLayered on classic scanning.
4.5 lockedTuning custom instructionsWhen prompt vs agent vs instructions.
4.6 lockedHooks: types & importanceBeta-taker priority topic.

Phase E · Orchestrate multi-agent coordination

— / — 15–20%

The 4-role canonical taxonomy. Distinctive exam content.

E0 lockedWorktrees + concurrencyClassic git multi-checkout.
5.1 locked4-role taxonomyOrchestrator · Planner · Coder · Designer.
5.2 lockedPer-agent model selectionClaude planning, GPT coding, Gemini design.
5.3 lockedDelegation patternsSubagent spawning.
5.4 lockedFleet modeParallel agent spawn.
5.5 lockedWorktree conflictsWhen agents collide.
5.6 lockedGitHub Skills disambiguationNot the same as Anthropic skills.

Phase F · Implement guardrails and accountability

— / — 10–15%
F0 lockedRulesets, CODEOWNERS, audit logClassic governance surface.
6.1 lockedWhat "guardrails" coverPolicy vs permission vs runtime.
6.2 lockedRulesets as agent guardrailsBranch protection for agents.
6.3 lockedHooks for guardrailsPre/post tool-use hooks.
6.4 lockedAudit log events for agentsMemorize exact event names.
6.5 lockedMCP allow-list enforcementConfig + audit trail.
6.6 lockedRisks & mitigations playbookCloud agent risk matrix.
6.7 locked"Agents propose; humans accept"Verbatim, likely MCQ stem.

Phase G · Drill & validate

— / —
G.1 lockedMock exam #1Full 6-domain simulation.
G.2 lockedWeak-domain re-teachAnywhere < 800.
G.3 lockedFlashcard drillYAML keys, event names, flags.
G.4 lockedMock exam #2Full, timed. Must clear 800.

Phase H · Exam day

— / —
H.1 lockedLast 24h flashcard passNo new concepts. Sleep.
H.2 lockedPre-exam checkID, rig, check-in window.
H.3 lockedSit examWhen you're ready.

Sources

Every exam-specific fact in every lesson is cited. Classic GitHub content uses any solid source. Agent-touched content is Tier-1 only and refetched if older than 90 days.

About

An independent, self-study tool for the GitHub GH-600 exam. Lessons are written one at a time, just-in-time, as the learner advances. No marketing, no funnel, no email capture. Sources cited or it doesn't ship.

Pass mark: 700 / 1000 · self-paced · free for everyone.

Unofficial. Not affiliated with, authorized, endorsed by, or sponsored by GitHub, Inc. or Microsoft Corporation. “GitHub”, “GitHub Copilot”, and “GH-600” are trademarks of their respective owners, referenced here for identification and study purposes only. This is not an official study guide; for the official guide see Microsoft Learn.