Skip to content

docs: optimize slides.md for readability and presentation crispness #21837

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
pelikhan merged 3 commits into from
Mar 19, 2026
Merged

docs: optimize slides.md for readability and presentation crispness #21837

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
d00d0c5
Initial plan
Copilot Mar 19, 2026
823f619
docs: optimize slides.md for readability and presentation crispness
Copilot Mar 19, 2026
6ba76ab
Merge branch 'main' into copilot/optimize-slides-for-readability
pelikhan Mar 19, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
159 changes: 38 additions & 121 deletions docs/slides/index.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@ mermaid.initialize({ startOnLoad: true });
| Runs CI/CD pipelines | Orchestrates AI agents |
| Debugging code | Debugging agent behavior |

> The role evolves: from coding to orchestrating intelligent agents
> The role evolves: from coding to orchestrating agents

---

Expand All @@ -40,7 +40,7 @@ mermaid.initialize({ startOnLoad: true });
- **Supervisor** — Monitor running agents and handle exceptions
- **Debugger** — Diagnose workflow behavior and improve prompts

> The human stays in the loop while AI handles execution
> Human oversight, AI execution

---

Expand Down Expand Up @@ -73,17 +73,12 @@ flowchart LR

# Continuous Integration to Continuous AI

- **Accessibility review** - Automated WCAG compliance checks

- **Documentation** - Auto-generate API docs and README files

- **Code review** - AI-powered PR analysis and suggestions

- **Test improvement** - Identify missing test coverage

- **Bundle analysis** - Monitor package size and dependencies

- **Issue triage** - Automated labeling and prioritization
- **Accessibility review** — Automated WCAG compliance checks
- **Documentation** — Auto-generate API docs and README files
- **Code review** — AI-powered PR analysis and suggestions
- **Test improvement** — Identify missing test coverage
- **Bundle analysis** — Monitor package size and dependencies
- **Issue triage** — Automated labeling and prioritization

> <https://githubnext.com/projects/continuous-ai/>

Expand All @@ -95,8 +90,6 @@ https://github.com/github/gh-aw/issues/1920

# Evolution: LLMs to SWE Agents

## From code completion to autonomous workflows

**2021: GitHub Copilot** - AI-powered code completion

**2022: ChatGPT** - Conversational AI assistant
Expand All @@ -111,7 +104,7 @@ https://github.com/github/gh-aw/issues/1920

# CI/CD with GitHub Actions

YAML workflows as configuration stored in `.github/workflows/` that trigger on events like push, pull requests, issues.
YAML workflows stored in `.github/workflows/`, triggered on events like push, pull requests, or issues.

```yaml
on:
Expand All @@ -129,7 +122,7 @@ jobs:

# The "Lethal Trifecta" for AI Agents

AI agents become risky when they combine **three capabilities** at once:
AI agents become dangerous when these **three capabilities** combine:

- **Private data access**

Expand All @@ -145,6 +138,8 @@ AI agents become risky when they combine **three capabilities** at once:

## Safe by design. Useful by default.

> The best developer tools protect you from catastrophe while letting you build something real

---

# From Scratch to MakeCode
Expand All @@ -157,7 +152,7 @@ AI agents become risky when they combine **three capabilities** at once:

These environments share one superpower:

> You're protected from catastrophe — and you can still build something **real and delightful**
> Protected from catastrophe — still building something **real and delightful**

---

Expand Down Expand Up @@ -232,7 +227,7 @@ Summarize issue and respond in a comment.

---

# Compiled Action Yaml
# Compiled Action YAML

```yaml
jobs:
Expand Down Expand Up @@ -284,7 +279,7 @@ network:
tools:
web-fetch:
---
Fetch latest TypeScript docs report findings in a comment.
Fetch latest TypeScript docs and report findings in a comment.
```

> Control external access for security
Expand Down Expand Up @@ -363,8 +358,6 @@ mcp-servers:

# Monitoring & Optimization

Track workflow performance and AI agent behavior.

```sh
# View recent runs
gh aw logs
Expand Down Expand Up @@ -404,8 +397,6 @@ Review this PR with context from previous reviews:

# Playwright + Upload Assets

## Browser automation for web app testing

```yaml
---
on:
Expand All @@ -418,11 +409,10 @@ safe-outputs:
upload-asset: # Attach screenshots to artifacts
---
Test the web application:
1. Navigate to the deployed preview URL
1. Navigate to the preview URL
2. Take screenshots of key pages
3. Check for visual regressions
4. Validate responsive design (mobile, tablet, desktop)
5. Create issue with findings and screenshots
3. Check for visual regressions and responsive design
4. Create issue with findings and screenshots
```

**Use cases:** Visual regression, accessibility audits, E2E validation for SPAs
Expand All @@ -447,11 +437,6 @@ safe-outputs:
# RECOMMENDED: Use sanitized context
Analyze this issue content (safely sanitized):
"${{ steps.sanitized.outputs.text }}"

Metadata:
- Issue #${{ github.event.issue.number }}
- Repository: ${{ github.repository }}
- Author: ${{ github.actor }}
```

**Auto-sanitization:** @mentions neutralized, bot triggers blocked, malicious URIs filtered
Expand All @@ -462,14 +447,10 @@ Metadata:

## Multi-layered defense in depth

GitHub Agentic Workflows implements a comprehensive security architecture with multiple isolation layers to protect against threats.

**Key Security Principles:**

- Container isolation for all components
- Network firewall controls at every layer
- Minimal permissions by default
- Separation of concerns
- Separation of concerns (agent, tools, outputs)

---

Expand Down Expand Up @@ -517,112 +498,48 @@ flowchart TB

# Security Layer 1: Coding Agent Sandbox

## Isolated agent process

**Coding Agent Sandbox:**

- Agent process runs in isolated container
- Read-only permissions by default
- No direct write access to repository
- Limited system access
**Agent sandbox** — isolated container, read-only by default, limited system access

**Primary Proxy/Firewall:**

- Filters outbound traffic from agent
- Controls access to MCP Gateway
- Enforces network allowlists
**Primary proxy/firewall** — filters outbound traffic, controls MCP Gateway access, enforces network allowlists

---

# Security Layer 2: MCP Gateway

## Central routing with access controls

**MCP Gateway (mcpg):**

- Central routing component
- Manages communication between agents and services
- Validates tool invocations
- Enforces permission boundaries

**Benefits:**
**MCP Gateway (mcpg)** — central routing between agents and services

- Single point of control
- Auditable tool access
- Prevents direct agent-to-service communication
- Validates tool invocations and enforces permission boundaries
- Single point of control — no direct agent-to-service access
- Full audit trail for tool calls

---

# Security Layer 3: Tool Sandboxes

## Isolated MCP servers and skills
**MCP servers & skills** — each runs in its own container, non-root, dropped capabilities

**MCP Server & Skill Sandboxes:**

- Each MCP server runs in own container
- Each skill runs in separate sandbox
- Non-root user IDs
- Dropped capabilities

**Secondary Proxy/Firewalls:**

- Additional proxy layers for egress traffic
- Domain-specific allowlists
- Defense against data exfiltration
**Secondary proxies** — egress filtering, domain allowlists, defense against data exfiltration

---

# Security Layer 4: Service Access

## Controlled external communication

**Service Layer:**

- External services accessed through proxies
- Multiple security controls before reaching services
- Comprehensive audit trail
- Network traffic monitoring
External services accessed only through proxies — multiple controls before reaching any service

**Defense in Depth:**
Even if one layer is compromised, multiple additional security controls remain in place.
> Defense in depth: if one layer is compromised, additional controls remain in place

---

# Security Features Summary

**Container Isolation:**

- GitHub Actions Jobs in VMs
- Separate sandboxes for agent, MCP servers, skills

**Network Controls:**

- Proxy/firewall at every layer
- Domain allowlisting
- Ecosystem-based controls (node, python, containers)

**Permissions:**

- Read-only by default
- Safe outputs for write operations
- Explicit permission grants

**Supply Chain Integrity:**

- Classic: GitHub Actions pinned to commit SHAs at compile time
- Agentic: Protected files guard CI/CD configuration from tampering

**GitHub Guard:**

- `min-integrity` restricts agent access by content trust level
- Secrecy & integrity metadata on safe output messages
- Automatic `min-integrity: approved` for public repositories

**Monitoring:**

- Threat detection
- Audit logs
- Workflow run analysis
| Layer | Protection |
|---|---|
| **Containers** | VMs + sandboxes for agent, MCP servers, skills |
| **Network** | Proxy/firewall at every layer, domain allowlisting |
| **Permissions** | Read-only default, safe outputs for writes |
| **Supply Chain** | Pinned action SHAs, protected CI/CD files |
| **GitHub Guard** | `min-integrity`, secrecy & integrity metadata |
| **Monitoring** | Threat detection, audit logs, run analysis |

---

Expand Down Expand Up @@ -662,7 +579,7 @@ Analyze issue and create implementation PR

**Visit:** <https://github.github.com/gh-aw/introduction/architecture/>

Security is foundational to GitHub Agentic Workflows. We continuously evolve our security controls and welcome community feedback.
---

# Getting Started (Agentically)

Expand Down
Loading