Skip to content

Latest commit

 

History

History

copilot-cli

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 

README.md

status active
updated 2026-11-03
section Agents at Your Keyboard
audience
developer
platform-engineer
level applied
duration 45
format core-talk
decision Where can I put AI help so decisive evidence and the right decision maker stay within reach?
prerequisites
surfaces
related
copilot-web
copilot-azure-mcp
references
url label verified
Install Copilot CLI
2026-08-25
url label verified
Plan before you build; steer as you go
2026-08-25
url label verified
Add models and discover Ollama
2026-11-02
url label verified
Allowing safe commands
2026-11-02
url label verified
Managed settings enforcement
2026-11-02
url label verified
Dynamic workflows
2026-11-02
url label verified
Dynamic workflows public preview
2026-11-02
url label verified
Parallel tasks with /fleet
2026-08-25
url label verified
Steer a session remotely
2026-11-02
url label verified
Use session data with /chronicle
2026-08-25
url label verified
Add repository instructions
2026-08-25
url label verified
CLI best practices
2026-08-25
url label verified
Official CLI releases
2026-11-02
url label verified
Stable v1.0.105
2026-11-02

GitHub Copilot CLI: Intelligence Within Reach

The Question This Talk Answers: "How do I get AI help right where I’m working?"

Duration: 45 minutes | Audience: Developers and platform engineers working with repository, CI, and running-system evidence

Thesis: Bring intelligence to the work, and keep the worker within reach.

Present and Study

  • Open the deck: 36 slides (25 teaching, 11 spine); estimated 40:00 talk plus 5:00 for questions. Press N for speaker notes.
  • Speaker notes and handout: Complete talk track, proposed live checks, source evidence, boundaries, and recovery paths.

The native teaching source owns the talk. Product evidence was reviewed November 2, 2026; stable baseline v1.0.105. The distance-first narrative was approved November 3. Representative traces are scripts, not observed results.

Why use CLI here? It can minimize the distance between the person, the evidence, the process, and the next useful moment. This is a practitioner thesis, not a historical origin claim or a ranking of other surfaces.

1. Worker ↔ Work

Keep the developer’s request, Docker tool outputs, and next decision in one inspectable loop. Clarify the task, review a scoped plan, and use the listener and valid 5001:5000 mapping to choose a proposed HTTP probe. No successful repair is claimed.

2. Intelligence ↔ Data

Start Copilot on the authorized staging host so tools can reach decisive evidence. Inspect inference placement separately, compare a bounded log finding with its synthetic source, and choose cloud deliberately for repository work. Host-local tools do not imply host-local inference or zero data movement.

3. Workflow ↔ Running System

Let failed CI run copilot -p and return a proposed why beside the original red test. Save recurring staging investigation as a public-preview dynamic workflow: understand the process, describe and review its creation, confirm registration, then invoke it and inspect the report. Permissions and containment bound the actions close to the system.

4. Now ↔ Later

Carry forward the right thing: a reusable procedure, delegated task, scheduled intent, checked context, reviewed guidance, recoverable edits, or continuing session. Experimental schedules require an active session. Original-host steering requires a running host and process; connected cloud continuation preserves different state. Verify identity, scope, files, and actual test evidence.

What You Can Do Today

Worker ↔ Work

  • Ask for two read-only checks beside a runtime task; inspect their outputs and choose the next probe.

Intelligence ↔ Data

  • Analyze an approved synthetic source on its host; inspect the provider and compare the finding with source.

Workflow ↔ Running System

  • Return one bounded CI failure explanation; compare why.md with the log and keep the original failed test status.

Now ↔ Later

  • Continue a low-risk task; verify the session, retained scope, files, and actual next check. Confirm access and current billing before cloud work.

Apply It to Your Work

  • Candidate task: Choose one distance in a real task; capabilities can close several distances.
  • Decisive context: Identify the host, files, service state, and provider allowed to receive context.
  • Delegation and authority: Name the permission decision maker and the human who accepts the result.
  • Evidence: Define an artifact and a signal that makes you keep, adjust, or stop the pattern.
  • Boundary: Stop when source evidence disagrees or access, data handling, or acceptance ownership is unclear. Preserve the artifact and re-scope with the owner.

Key Artifacts

  • Walkthrough and proposed checks: Safe rehearsal paths for planning, CI, model placement, permissions, containment, workflows, and both continuity paths.
  • Illustrative workflow extension: The kind of extension.mjs Copilot writes for the staging investigation, using the API in the guidance bundled with CLI v1.0.105; syntax-checked and stub-SDK tested, not a recorded run. and fixture inputs: Dependency-free Node.js fixture checks; a Copilot extension adapter requires separate authoring and audit.
  • Teaching source: Slide data, full notes, source map, and timing.
  • Approved recipe: Structural rationale and actual per-slide plan.

Related Patterns