Table of Contents

Build two complete AI collaboration workflows for developers, operations staff, and non-coding contributors. Carry one synthetic retention change through GitHub-first and GitHub with Confluence Cloud/Jira Cloud. Each track produces source records, review evidence, publication history, recovery records, and a fresh-session handoff.

What You Will Build

The fictional project is export-service-lab. Approved retention starts at seven days. Your proposal changes it to thirty days. No production data or deletion job is involved. The exercise tests authority and coordination, not compliance requirements.

TrackAuthoritative homesFinal artifact
GitHub-firstRepository policy, requirements, implementation, runbookReviewed PR, consistency log, Issue, handoff
Mixed workplaceGitHub implementation, Confluence requirement/runbook, Jira deliveryFixed proposal and cross-system ledger

This course implements the AI collaboration framework . The framework explains ownership. These lessons supply exact procedures, copyable policies, worked examples, verification, negative tests, troubleshooting, rollback, and exercises. Course policy choices are not vendor defaults.

Before You Begin

Who: contributor, product owner, operations owner, maintainer, and publisher. Assign people privately. Use separate people or organization-approved test identities for denial tests. Do not create misleading extra accounts to simulate independent review.

What and where: disposable GitHub repository, an approved coding/chat tool if desired, and dedicated Confluence/Jira Cloud sandboxes for Mixed. Python 3.10 or later runs the local lab without packages. Browser-only contributors ask the maintainer to run checks.

When and why: begin before attaching write-capable integrations. Reserve roughly 10 to 14 hours plus review across a one-week pilot. Finish GitHub-first before repeating the change in Mixed. The goal is reconstructable reviewed delivery, not agreement between two assistants.

Plan boundaries: GitHub protection depends on plan/visibility. Confluence Free lacks content restrictions. Jira Free lacks configurable permission schemes. Setup lessons explain the limits and provide official references. No particular paid chat connector, hosted coding agent, Rovo feature, or marketplace app is required.

Readiness check: open a GitHub repository, identify its default branch, explain what a PR changes, and read a small JSON record such as {"retention_days": 7}. For local checks, confirm Python and Git run in your terminal. For browser-only work, complete the foundation, repository, and browser lessons with a maintainer. You do not need to run the agent and Actions lesson to submit an Issue.

First-time setup bridge: Create a GitHub account under your organization’s policy and ask a maintainer to create a disposable repository. Install Git from git-scm.com and Python 3.10 or later from python.org . Open Terminal on macOS or Linux, or PowerShell on Windows. Run git --version and python3 --version on macOS or Linux, or git --version and py -3 --version on Windows. Each command should print a version. If a command is missing, install the tool before the local track. GitHub’s authentication guide explains the supported sign-in paths. Do not place passwords or tokens in a repository.

Beginner environment check: Use a disposable personal computer or a course-provided lab workstation, not a production server. Save the ZIP in a folder you own and extract it into a new empty folder with the file manager. Open the extracted folder and confirm check.py, test_check.py, and baseline/ appear together. In macOS or Linux Terminal, change to that folder with cd and verify pwd and python3 --version. In Windows PowerShell, use Set-Location to enter the folder and verify Get-Location and py -3 --version. Copy the folder path from the file manager when typing it is error prone. If Python is missing, install it from the linked official installer and reopen the terminal. If Git is missing, install it from the linked official download and reopen the terminal. If the extracted files are nested one folder deeper than expected, enter the folder containing check.py before running a lesson command. Do not run the lab from the ZIP preview or a protected system directory. A learner without permission to install tools uses the browser Issue route with a maintainer and records the local checks as Blocked.

Harmless Git rehearsal: Create a disposable folder in your file manager and open a terminal there. Run git init. Save a one-line practice.txt with a text editor. Run git add practice.txt, then git commit -m "Practice commit". If Git asks for an author identity, run git config user.name "Your approved name" and git config user.email "[email protected]" inside this rehearsal repository. Replace both examples with your approved account details, then retry the commit. These commands set identity only for this repository. git status should report a clean working tree. This local rehearsal publishes nothing. The repository lesson covers the remote route and its verification.

Terms You Need First

TermMeaning in this course
AuthorityThe named record whose owner decides a specific fact, such as approved retention
Protected branchA GitHub branch with rules governing direct changes and merges
Pull request (PR)A proposed branch change for review before merging
Local checkoutThe repository files on your computer
RemoteThe hosted repository location connected to a local checkout
BranchA named line of commits used to isolate proposed changes
CommitA recorded snapshot of selected changes with an identifier
ManifestA file recording source revisions and hashes used for a check
Revision-bound reviewApproval of a specific proposal version and the source versions behind it
Read-backOpening published records again to confirm their final values
Retrieval-augmented generation (RAG)Supplying selected source text to an AI response at request time
Embedding and vector databaseA numeric text representation and a store used to find similar text. Both are optional here

Use the GitHub-first route when workplace access is unavailable. Finishing modules 1 to 4 and the GitHub-first capstone produces a partial course result. The full two-track completion standard requires the Mixed workplace exercises.

RouteAccess to arrange before starting
GitHub-firstA disposable repository, a maintainer, separate contributor and reviewer accounts, and branch protection available for its plan and visibility
Mixed workplaceGitHub-first access plus Confluence and company-managed Jira Cloud sandboxes, setup administrator rights, separate contributor and owner accounts, and plans supporting the permission controls in modules 5 and 6

Check the plan controls in the setup lessons before inviting learners. If a control is unavailable, label its live test Blocked and complete the GitHub-first route. A simulated denial does not replace a platform denial.

Course Modules

#LessonReader artifact
1Pilot FoundationCharter, authority register, policy, test matrix
2GitHub Repository SetupSource map, owners, protected boundary
3Agent Adapters and ActionsAdapters, manifest, trusted-base check
4Browser ContributionsIssue, reviewed PR, non-coder handoff
5Confluence and Jira SetupAuthority pages and permission evidence
6Cross-System PublicationFixed proposal and publication ledger
7Map-First RetrievalLookup procedure and optional RAG evaluation
8Drift, Recovery, and HandoffsRecovery decision and fresh verification
9Retention CapstoneTwo runs and independent acceptance

How to Use It

  1. Approve the foundation before connecting tools.
  2. Complete modules 2 to 4 for coding-agent and browser/chat GitHub contribution.
  3. Complete modules 5 and 6 to preserve workplace authority across systems.
  4. Run modules 7 and 8 for discovery, drift, denial, recovery, and handoff.
  5. Deliver the capstone in both tracks and obtain independent review.

Keep each artifact for the next module. Lessons explain who acts, what changes, when to proceed, where evidence lives, why the control matters, and how to verify it. Worked examples are illustrative. Expected outcomes become observations only after your sandbox produces evidence.

Download the Lab

The synthetic lab archive contains baseline records, a Python validator, a runbook, and ten unit tests. Extract into an empty directory and follow module 1. It makes no network calls and requires no credentials.

The validator is not an approval service. It detects source-byte drift and inconsistent requirement, configuration, proposal, and runbook records. Live permissions, identity, owner approval, and external publication require separate course tests.

Completion Standard

CaseRequired demonstration
Approved changeReview and consistent publication read-back
Stale contextChanged base rejected
Unauthorized writeContributor denied by platform controls
ConflictAuthority owner reconciles sources
Partial publicationIncomplete delivery remains visible
Access denialProtected text absent from unauthorized task
RecoveryReviewed completion or compensation
HandoffFresh session independently reads sources

Completion requires observed evidence. Mark missing controls Blocked. Synthetic success supports another bounded pilot, not automatic production authorization.

Example evidence row, illustrative only: GitHub-first unauthorized-write test. Actor role: contributor. Initial main revision: record the sandbox commit. Attempt: direct push with a harmless synthetic edit. Expected: server rejection. Observed: enter the actual terminal response or mark Not run. Final revision: read main again and record its commit. Evidence: github-first/03-unauthorized-write.md. Reviewer: independent maintainer. Do not label this row Passed until the sandbox supplies the response and unchanged revision.

Next Steps

Start with Pilot Foundation . After the capstone, review provider handling, runtime tests, secrets, deployment, and production access before selecting another pilot. Return to Courses and Playbooks for related study.