Skip to content

Instantly share code, notes, and snippets.

@johnlindquist
Created December 7, 2025 16:08
Show Gist options
  • Select an option

  • Save johnlindquist/b2c9ddbdc90fdedf32a66099e605ca53 to your computer and use it in GitHub Desktop.

Select an option

Save johnlindquist/b2c9ddbdc90fdedf32a66099e605ca53 to your computer and use it in GitHub Desktop.
Microsoft AI Dev Days Talk Outline: Supercharging GitHub Copilot CLI with markdown-agent

Markdown as the New Programming Language: Supercharging GitHub Copilot CLI

Microsoft AI Dev Days Talk Outline Duration: 15-20 minutes + demos


The Hook (1 minute)

"What if I told you the most powerful programming language of 2025 isn't TypeScript, Python, or Rust—it's Markdown?"

Opening Statement:

  • Software has fundamentally shifted from code that does things to instructions that orchestrate AI
  • The skill that matters now: not learning tools, but creating instructions
  • GitHub Copilot CLI is the bridge—and we're about to supercharge it

Section 1: The Shift in Software Development (3 minutes)

Key Points:

  1. The Old Model: Developers write code → Tools execute code
  2. The New Model: Developers write instructions → AI interprets → Tools orchestrate
  3. Why This Matters for Microsoft Developers:
    • Azure DevOps pipelines are already YAML (instructions!)
    • GitHub Actions are already declarative
    • Copilot is already reading your comments and docs
    • You're already halfway there

The Insight:

"Every README you've ever written was training you for this moment."


Section 2: GitHub Copilot CLI—The Unsung Hero (4 minutes)

Key Points:

  1. Non-Interactive Mode is the Game Changer

    gh copilot suggest -t shell "find all TypeScript files modified today" --no-interactive
    • No prompts, no confirmations—pure automation
    • Perfect for CI/CD pipelines
    • Enterprise-grade: audit trails, policy compliance
  2. Three Modes, Three Superpowers:

    • suggest -t shell — Shell commands
    • suggest -t gh — GitHub CLI operations
    • suggest -t git — Git operations
  3. Enterprise-Friendly by Design:

    • Runs through your existing GitHub auth
    • Respects organization policies
    • No external API keys needed
    • Audit logs for compliance

Demo Idea #1: "The One-Liner"

# Old way: Google → StackOverflow → copy → paste → debug
# New way:
gh copilot suggest -t shell "recursively find all TODO comments in .cs files" --no-interactive | sh

Section 3: Enter markdown-agent—Instructions as Files (5 minutes)

The Problem:

  • Copilot CLI is powerful but ephemeral
  • Prompts disappear after execution
  • No version control, no sharing, no iteration

The Solution: ma (markdown-agent)

  • Markdown files ARE your agents
  • YAML frontmatter = configuration
  • Body = instructions
  • Filename = command (task.copilot.md → runs copilot)

Demo Idea #2: "Your First Agent File"

<!-- review.copilot.md -->
---
command: gh copilot suggest
type: shell
---

Analyze the git diff for the current branch and suggest:
1. Potential bugs
2. Missing test cases
3. Documentation gaps

Focus on: {{ focus | default: "security" }}

Run it:

ma review.copilot.md --focus performance

Key Points:

  1. Version Control Your Prompts — Treat instructions like code
  2. Share with Your Team — PR your agent files
  3. Template Variables — LiquidJS powers dynamic prompts
  4. Import System@./src/**/*.cs pulls in context

Section 4: The Escalation Pattern (4 minutes)

The Concept: Intelligent Scripts That Know When They Need Help

Tier 1: Copilot CLI (fast, cheap, scoped)

  • Quick suggestions
  • Shell command generation
  • Git operations

Tier 2: Codex/GPT-4 (more context, deeper analysis)

  • Multi-file refactoring
  • Architecture decisions
  • Complex debugging

Tier 3: Claude/Opus (maximum reasoning)

  • System design
  • Security audits
  • Production incident analysis

Demo Idea #3: "The Escalating Pipeline"

<!-- analyze.copilot.md -->
---
command: gh copilot suggest
type: shell
---

Analyze this error log and suggest a fix:
@./logs/error.log

If the issue involves multiple services, output: ESCALATE
# The intelligent script
result=$(ma analyze.copilot.md)
if [[ "$result" == *"ESCALATE"* ]]; then
  ma deep-analyze.claude.md  # Escalate to more powerful agent
fi

Why This Matters for Enterprise:

  • Cost Control — Use the cheapest capable model
  • Speed — Fast answers stay fast
  • Compliance — Copilot stays in your GitHub ecosystem
  • Flexibility — Escalate when needed

Section 5: Real-World Microsoft Developer Scenarios (3 minutes)

Scenario 1: PR Review Pipeline

<!-- pr-review.copilot.md -->
---
command: gh copilot suggest
type: gh
---

Review PR #{{ pr_number }} for:
- Breaking changes to public APIs
- Missing XML documentation
- Azure resource configuration issues

Scenario 2: Azure Deployment Assistant

<!-- deploy-check.copilot.md -->
---
command: gh copilot suggest
type: shell
---

Before deploying to {{ environment | default: "staging" }}:
1. Check if all required Azure resources exist
2. Validate ARM template syntax
3. List any secrets that will be rotated

@./infrastructure/main.bicep

Scenario 3: .NET Migration Helper

<!-- migrate-check.copilot.md -->
---
command: gh copilot suggest
type: shell
---

Analyze this .NET Framework project and identify:
1. NuGet packages that need updating for .NET 8
2. API calls that have changed
3. Configuration patterns that need migration

@./src/**/*.csproj
@./src/**/*.config

Section 6: The Bigger Picture (2 minutes)

Key Takeaways:

  1. Markdown is Infrastructure

    • Your prompts are now deployable artifacts
    • Version them, test them, ship them
  2. Copilot CLI is Your Fast Path

    • Non-interactive mode unlocks automation
    • Enterprise-ready out of the box
    • Escalation keeps you flexible
  3. The New Developer Skill

    • Writing code → Writing instructions
    • Using tools → Orchestrating agents
    • Solo work → Human-AI collaboration

The Call to Action:

"Start with one markdown file. One agent. One workflow you automate this week."


Closing (1 minute)

The Memorable Line:

"In 2020, we said 'every company is a software company.' In 2025, every software developer is an instruction architect. And the best architecture starts with a simple .md file."

Resources:

  • markdown-agent: npm install -g markdown-agent
  • GitHub Copilot CLI: gh extension install github/gh-copilot
  • Today's demos: [gist link]

Bonus Demo Ideas (if time permits)

"The Context-Aware Agent"

<!-- smart-fix.copilot.md -->
---
command: gh copilot suggest
type: shell
inputs:
  - name: issue
    prompt: "Paste the error message:"
---

Given this error:
{{ issue }}

And the recent changes:
!`git diff HEAD~3`

Suggest a targeted fix.

"The Team Knowledge Base"

# Every developer gets the same quality prompts
git clone [email protected]:myorg/team-agents.git ~/.agents
ma ~/.agents/code-review.copilot.md
ma ~/.agents/security-scan.copilot.md
ma ~/.agents/perf-check.copilot.md

Talk Flow Summary

Time Section Energy
0:00-1:00 Hook High
1:00-4:00 The Shift Building
4:00-8:00 Copilot CLI Deep Dive + Demo High
8:00-13:00 markdown-agent + Demos High
13:00-17:00 Escalation + Scenarios Practical
17:00-19:00 Bigger Picture Inspiring
19:00-20:00 Closing Memorable

Talk outline created for Microsoft AI Dev Days Theme: Software is now instructions + tools

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment