Skip to main content

gh300-lab-creator

Create short GH-300 practice exercises (10-20 minutes) that are executable and self-validating. Every exercise includes prerequisites, exact tasks, validation steps, expected outcomes, and rollback. Use when the user asks for a hands-on exercise, practice flow, or guided walkthrough.

Quellinformationen

Repository
timothywarner-org/copilot-cert-prep
Letzte Quellaktivität
23. September 2026 um 14:13
Erkannte Sprache von SKILL.md
Englisch
Sterne
97
Forks
71

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

Datei-Explorer
5 Dateien

SKILL.md wird angezeigt

SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
gh300-lab-creator
description
Create short GH-300 practice exercises (10-20 minutes) that are executable and self-validating. Every exercise includes prerequisites, exact tasks, validation steps, expected outcomes, and rollback. Use when the user asks for a hands-on exercise, practice flow, or guided walkthrough.
# Skill: gh300.practice_exercises.micro.validated **Description:** Create short GH-300 practice exercises (10-20 minutes) that are executable and self-validating. ## Grounding **Required sources:** - Microsoft Learn (primary truth source for feature and policy behavior; access via the **Microsoft Learn MCP server**) - Microsoft Learn code samples for syntax when applicable ## Guardrails - Keep the exercise within GH-300 scope. - Prefer low-risk actions that can run in a local repo or demo org. - No contractions. - Always include rollback or cleanup steps. - Always use current terminology. ## Bundled assets Use the skill-local bundle to keep outputs consistent and auditable. - `resources/source-pack.md` - `resources/microsoft-voice-principles.md` - `resources/lab-quality-checklist.md` - `scripts/validate-output.js` ## Fictional company randomization (non-negotiable) Use fictional company names from `references/fictional-companies.md` for any scenario context. Randomize the company selection. ## Timebox guidance An exercise should contain no more than 12 steps total across all tasks. If content exceeds that, split into two exercises. ## Workflow 1. Choose a single GH-300 objective from `references/gh300-objectives.md` and state it at the top. 2. Ground the intended behavior in Microsoft Learn. 3. Draft steps using one primary surface (VS Code, GitHub.com, or CLI). 4. Verify command syntax if commands are included. 5. Add validation gates after each major step. 6. Add rollback that exactly reverses the changes. 7. Run the checks in `resources/lab-quality-checklist.md` and apply `scripts/validate-output.js` rules before final delivery. ## Output format ```yaml exercise: title: "<Action + focus area>" objective: "<One sentence outcome tied to GH-300>" skill_area: "<GH-300 skill area>" estimated_time: "<10-20 min>" prerequisites: - "<Workspace and account requirements>" starting_state: - "<What must already exist>" tasks: - name: "<Task 1 name>" steps: | <Numbered steps only when sequencing matters.> validation: - "<Validation action + expected result>" - name: "<Task 2 name>" steps: | <...> validation: - "<...>" troubleshooting: - symptom: "<common failure>" fix: "<precise fix>" cleanup: steps: | <Exact rollback steps> validation: - "<Check that rollback succeeded>" references: - "<Microsoft Learn URL(s)>" ``` ## Delivery rules Exercises are delivered in full in a single message. If multiple exercises are requested, deliver each exercise sequentially in the same message.
Auf GitHub ansehen