Git Ape Onboarding

Git-Ape Onboarding

Use this skill to bootstrap a repository for Git-Ape deployments by executing the onboarding workflow directly from Copilot Chat.

This skill is the source of truth for onboarding behavior. Do not depend on a standalone repository script for setup logic.

Onboarding Modes

This skill operates in two independent modes:

• Repository CI/CD onboarding (default). Configures one repository + subscription(s) for Git-Ape deployments: OIDC, federated credentials, RBAC, GitHub environments, secrets, and scaffolded workflows. This is the bulk of the skill (see Command Playbook).
• Enterprise distribution (.github-private). Rolls Git-Ape out to every user on your org/enterprise Copilot plan by scaffolding a .github-private repo with managed-settings.json plugin standards (and an optional agents/ directory). See Mode: Enterprise Distribution.

The two modes are complementary, not alternatives: enterprise distribution installs the tooling for everyone, while repository onboarding wires up Azure access for a specific repo. A fully onboarded user typically needs both.

When to Use

• First-time setup of a repository for Git-Ape
• New subscription onboarding (single environment)
• Multi-environment onboarding (dev/staging/prod across different subscriptions)
• New user handoff where OIDC, RBAC, and GitHub environments must be created
• Enterprise-wide distribution: rolling Git-Ape out to every user on your org/enterprise Copilot plan via a .github-private repo, so the plugin (agents + skills + azure-mcp) auto-installs on authentication — no per-user gh plugin install required

DO NOT USE FOR: re-deploying an already-onboarded repo (use git-ape), rotating or updating an existing secret or federated credential, drift detection setup alone (that is an optional sub-step covered by Step 10), or general Azure resource deployment.

What It Configures

This skill configures:

1. Entra ID App Registration and service principal (or reuses existing)
2. OIDC federated credentials for GitHub Actions
3. RBAC role assignment(s) on subscription scope
4. GitHub environments (azure-deploy*, azure-destroy)
5. Required GitHub secrets (AZURE_CLIENT_ID, AZURE_TENANT_ID) and the AZURE_SUBSCRIPTION_ID variable
6. Scaffolded GitHub Actions workflow files (git-ape-plan.yml, -deploy.yml, -destroy.yml, -verify.yml, -drift.{md,lock.yml}) and deployment standards (.github/copilot-instructions.md) into the user's working copy
7. (Optional) The COPILOT_GITHUB_TOKEN repository secret that powers the agentic drift-detection workflow (git-ape-drift.lock.yml) — only when the user opts into scheduled drift detection

Prerequisites

Before onboarding, run the prereq-check skill to verify all required tools are installed and auth sessions are active:

/prereq-check

The prereq-check skill validates: az (≥ 2.50), gh (≥ 2.0), jq (≥ 1.6), git, and active Azure/GitHub auth sessions. If anything is missing, it shows platform-specific install commands.

Do NOT proceed with onboarding until prereq-check reports ✅ READY.

Additionally, the Azure identity used must have Owner or User Access Administrator on the target subscription(s), and the GitHub identity must have admin access to the target repository.

Invariants

These rules are non-negotiable. The agent MUST NOT improvise around them.

• Default branch is always main. Never use master, never auto-detect a non-main default, and never substitute any other name. All federated credential subjects, environment branch policies, and example commands use refs/heads/main / the literal string main. If a user's repository uses something other than main, prompt for it once and use the user-supplied value explicitly — never silently default to master.
• Federated credential names use the fc-main-branch form, not fc-master-branch. See Step 5 for the canonical subject strings.
• Workflows ship main-targeted triggers. The scaffold step copies workflow files that reference branches: [main]; do not rewrite them to master.

Execution Modes

Interactive (recommended for first-time use)

Invoke the skill from chat and let the agent gather missing parameters:

/git-ape-onboarding

Parameterized single environment

/git-ape-onboarding onboard https://github.com/org/repo on subscription 00000000-0000-0000-0000-000000000000 with Contributor

Parameterized multi-environment

/git-ape-onboarding onboard https://github.com/org/repo with dev on 11111111-1111-1111-1111-111111111111 as Contributor, staging on 22222222-2222-2222-2222-222222222222 as Contributor, prod on 33333333-3333-3333-3333-333333333333 as Contributor+UserAccessAdministrator

Enterprise distribution (.github-private)

Invoke the skill in enterprise mode to scaffold the org/enterprise distribution repo instead of onboarding a single deployment repo:

/git-ape-onboarding distribute git-ape to the <org> enterprise

This runs the enterprise distribution playbook rather than the repository CI/CD playbook below.

Command Playbook

When the agent executes this skill, it should run the equivalent Azure and GitHub CLI commands directly in this order:

1. Validate prerequisites and current auth context.
2. Resolve repo metadata:

gh repo view <org>/<repo>
gh api repos/<org>/<repo> --jq '{repo_id: .id, owner_id: .owner.id}'
gh api orgs/<org>/actions/oidc/customization/sub --jq '.use_default'

3. Create or reuse the Entra app registration and service principal:

CLIENT_ID=$(az ad app create --display-name "$SP_NAME" --query appId -o tsv)
az ad sp create --id "$CLIENT_ID"
TENANT_ID=$(az account show --query tenantId -o tsv)
OBJECT_ID=$(az ad app show --id "$CLIENT_ID" --query id -o tsv)

4. Build the OIDC subject prefix:

# default format
OIDC_PREFIX="repo:<org>/<repo>"

# if org customization returns false
OIDC_PREFIX="repository_owner_id:<OWNER_ID>:repository_id:<REPO_ID>"

5. Create federated credentials with these canonical subjects (always refs/heads/main — never master):
   • fc-main-branch subject "$OIDC_PREFIX:ref:refs/heads/main" description "Main branch deployments"
   • fc-pull-request subject "$OIDC_PREFIX:pull_request" description "Pull request plan/validate"
   • fc-azure-deploy subject "$OIDC_PREFIX:environment:azure-deploy" (one per environment in multi-env mode)
   • fc-azure-destroy subject "$OIDC_PREFIX:environment:azure-destroy"
6. Assign RBAC on each target subscription.
7. Set GitHub repo or environment secrets (AZURE_CLIENT_ID, AZURE_TENANT_ID) and the AZURE_SUBSCRIPTION_ID variable.
8. Create GitHub environments and branch policies when permissions allow.
9. Scaffold workflow files and deployment standards into the user's working copy (see below).
10. (Optional) Provision the drift detector engine credential (COPILOT_GITHUB_TOKEN) so the agentic drift workflow can run (see below).
11. Capture compliance and Azure Policy preferences (see below).
12. Verify federated credentials, role assignments, and secrets.

Step 9: Scaffold workflow files and deployment standards

The GitHub Actions workflows that power Git-Ape (git-ape-plan.yml, -deploy.yml, -destroy.yml, -verify.yml, -drift.md, -drift.lock.yml) and the deployment standards file (.github/copilot-instructions.md) ship as templates inside this skill at ./templates/.

After identity, secrets, and environments are configured, run the scaffold helper to copy these templates into the user's working copy. Two parity implementations ship — pick the one that matches the user's shell:

# macOS / Linux / WSL
./scripts/scaffold-repo.sh

# Windows (PowerShell 7+)
pwsh .github/skills/git-ape-onboarding/scripts/scaffold-repo.ps1

Both scripts produce byte-identical output and follow the same rules below. The onboarding-template-check workflow enforces parity on every PR.

The helper:

• Resolves the target repo root via git rev-parse --show-toplevel (override by passing an explicit path as the first argument).
• Copies each template only if the destination does not already exist (skip-with-notice on collision — never overwrites a customized file).
• Prints ✓ Created for new files, ⊝ Skipped for collisions, and a final Created N file(s), skipped M file(s). summary.
• Leaves all files unstaged. It does not run git add, git commit, git push, or open a pull request — the user decides how to land them.
• For each skipped file, prints a diff -u command pointing at the canonical template so the user can reconcile manually.

If the user already had a custom .github/copilot-instructions.md, the scaffold step skips it. Step 11 (below) handles that case explicitly.

Step 10: (Optional) Onboard the drift detector workflow

This step is optional. It is only needed if the user wants the scheduled drift-detection workflow (git-ape-drift.lock.yml) to run. The plan, deploy, destroy, and verify workflows do not depend on anything from this step — skip it entirely if the user is not enabling drift detection.

Unlike the other scaffolded workflows, git-ape-drift is a GitHub Agentic Workflow (authored with gh-aw) that runs on the GitHub Copilot engine. Its compiled .lock.yml opens with a hard preflight gate — "Validate COPILOT_GITHUB_TOKEN secret" — that fails the run immediately when the credential is missing. There is no fallback: the Azure OIDC secrets from Step 7 cover the workflow's deterministic pre-steps, but the agent itself needs its own engine token.

To onboard it:

1. Confirm intent. Ask the user whether they want scheduled drift detection. If not, skip this step.

2. Provision COPILOT_GITHUB_TOKEN as a repository secret — not an environment secret, because the daily schedule runs from main with no environment attached:

   gh secret set COPILOT_GITHUB_TOKEN --repo <org>/<repo>
   # paste the token when prompted — never pass it on the command line
   

   Token requirements:

   • A GitHub PAT (fine-grained or classic) belonging to an identity with an active GitHub Copilot seat.
   • The built-in GITHUB_TOKEN cannot drive the Copilot engine, so the token must be supplied explicitly.
   • The other gh-aw tokens (GH_AW_GITHUB_TOKEN, GH_AW_GITHUB_MCP_SERVER_TOKEN) are not required — they fall back to the auto-provided GITHUB_TOKEN.

3. (Only if recompiling.) The scaffolded .lock.yml runs as-is. The gh-aw CLI is needed only when the user edits git-ape-drift.md and wants to regenerate the lock file:

   gh extension install github/gh-aw
   gh aw compile
   

4. Smoke-test the workflow end to end:

   gh workflow run git-ape-drift.lock.yml --repo <org>/<repo>
   gh run list --workflow git-ape-drift.lock.yml --repo <org>/<repo> --limit 1
   

Never print the token value in chat output (see Safe-Execution Rules).

Step 11: Compliance & Azure Policy Preferences

After RBAC and environment setup, ask the user about compliance requirements and update the ## Compliance & Azure Policy section in .github/copilot-instructions.md:

1. Ask compliance framework:

   Which compliance framework should Git-Ape use for policy recommendations?
   - General Azure best practices (recommended)
   - CIS Azure Foundations v3.0
   - NIST SP 800-53 Rev 5
   - None — skip policy recommendations
   

2. Ask enforcement mode:

   How should policies be enforced initially?
   - Audit only (recommended — evaluate compliance without blocking)
   - Enforce (Deny — block non-compliant deployments immediately)
   

3. Update copilot-instructions.md with the user's choices:

   • If the file does not exist (scaffold step was skipped or scaffolding was not run), print the captured preferences in chat and ask the user to add them manually. Do NOT create a new file from scratch — that is the scaffold step's responsibility.
   • If the file exists AND contains a ## Compliance & Azure Policy section, edit the ### Compliance Frameworks and ### Policy Enforcement Mode subsections in place.
   • If the file exists but does NOT contain that section (user has a customized file), do NOT mutate it. Instead, print the captured preferences and a suggested patch in chat so the user can apply it.
   • In all cases, leave changes unstaged and let the user commit them.

Mode: Enterprise Distribution (.github-private)

Use this mode to distribute Git-Ape to everyone on an organization's or enterprise's Copilot plan at once, instead of onboarding one deployment repo. It scaffolds a special .github-private repository that GitHub Copilot reads to apply enterprise-managed plugin standards.

[!IMPORTANT] This mode configures tooling distribution only. It does not grant Azure access. Each user/repo that actually deploys still needs az login/OIDC and a per-repo run of the repository CI/CD playbook above.

Why the plugin route (not agents/ alone)

Git-Ape is a plugin that bundles agents + skills + the azure-mcp MCP server. The .github-private agents/ directory distributes standalone agents only — copying Git-Ape's agents there would ship them without their skills and MCP server, so they would load but fail. Distribute Git-Ape through managed-settings.json, which auto-installs the whole plugin.

[!NOTE] Standalone org/enterprise skills are "coming soon" per GitHub's docs. Today, Git-Ape's skills reach users because they are bundled in the plugin — already covered by the managed-settings.json route below.

What it configures

Scaffolds, into a .github-private repository working copy:

1. .github/copilot/managed-settings.json — registers the Azure/git-ape marketplace and enables the git-ape@git-ape plugin for all members.
2. README.md — governance, admin setup steps, and caveats for maintainers.
3. agents/.gitkeep — placeholder for optional standalone custom agents.

Prerequisites (enterprise mode)

• gh authenticated as a user with permission to create a repo in the target org (gh auth status).
• An enterprise owner to perform the AI-controls designation and ruleset steps (these are GitHub UI actions — see the hand-off below).
• The org that will own .github-private is part of the enterprise.

Enterprise distribution playbook

The agent can automate steps 1–4 via CLI; steps 5–6 are UI-only and must be handed off to an enterprise owner.

1. Confirm the target org/enterprise and ownership, then echo the plan and require explicit confirmation before creating anything.

2. Create (or reuse) the .github-private repo in the target org. --internal gives every enterprise member read access; use --private to grant access manually:

   gh repo create <org>/.github-private \
     --internal \
     --description "Copilot enterprise configuration (Git-Ape standards)"
   gh repo clone <org>/.github-private /tmp/github-private
   

3. Scaffold the canonical files into the cloned repo root. Two parity implementations ship — pick the one matching the user's shell, and pass the cloned repo path as the target:

   # macOS / Linux / WSL
   .github/skills/git-ape-onboarding/scripts/scaffold-enterprise.sh /tmp/github-private
   

   # Windows (PowerShell 7+)
   pwsh .github/skills/git-ape-onboarding/scripts/scaffold-enterprise.ps1 C:\path\to\github-private
   

   Both scripts produce byte-identical output and follow the same skip-with-notice / no-git rules as the repository scaffolder (Step 9 above). The onboarding-template-check workflow enforces parity on every PR.

4. Review and publish. Edit managed-settings.json if you want to also enable the optional ape-context@git-ape companion plugin, then review the README.md placeholders. The scaffolder leaves everything unstaged — let the user (or a reviewed PR) commit and push to the default branch:

   cd /tmp/github-private
   jq empty .github/copilot/managed-settings.json   # validate before publishing
   git add .github/copilot/managed-settings.json README.md agents/.gitkeep
   git commit -m "Add Git-Ape enterprise Copilot standards"
   git push
   

5. Hand off the enterprise designation (UI-only). Instruct an enterprise owner to open Enterprise → AI controls → Custom agents → Select organization and choose the org that owns .github-private. This same designation points the enterprise at the repo's managed-settings.json. There is no stable CLI/API for this during public preview — the agent must hand off with the link, not attempt to automate it.

6. (Recommended) Protect the files (UI-only). On the same AI-controls page, under "Protect agent files using rulesets", create a ruleset so only enterprise owners can merge changes.

Verification (enterprise mode)

# Confirm the standards landed on the default branch of the config repo
gh api repos/<org>/.github-private/contents/.github/copilot/managed-settings.json --jq '.path'

# Validate the published JSON is well-formed
gh api repos/<org>/.github-private/contents/.github/copilot/managed-settings.json \
  --jq '.content' | base64 --decode | jq empty && echo "✓ managed-settings.json is valid JSON"

Then, on a supported client (Copilot CLI, or VS Code 1.122+), a member of the designated org re-authenticates and confirms the git-ape plugin auto-installed. Users licensed by multiple billing entities must select this enterprise under "Usage billed to" in their personal Copilot settings.

After distribution: still onboard repos for Azure

Distribution installs the Git-Ape tooling everywhere, but deployments still need Azure identity. For each repository that will deploy, run the repository CI/CD playbook above (/git-ape-onboarding onboard <repo> ...) to wire up OIDC, RBAC, environments, and workflows.

Safe-Execution Rules

1. Echo target repository and subscription(s) before execution.
2. Require explicit user confirmation before running onboarding.
3. Never print secret values in chat output.
4. Summarize what was created or updated (app registration, federated credentials, role assignments, GitHub environments, scaffolded files).
5. If onboarding fails, surface the failing step and command context, then stop.
6. Never overwrite an existing .github/workflows/* file or .github/copilot-instructions.md. The scaffold helper enforces skip-with-notice; do not bypass it.
7. Never run git add, git commit, git push, or open a PR for the scaffolded files — leave them unstaged so the user decides how to land them.
8. Idempotency on re-run: If the skill is re-invoked after a partial failure, re-run from the last failing step — not from scratch. The Entra app, federated credentials, role assignments, and GitHub environments created before the failure are safe to reuse; do not create duplicates. Surface each already-provisioned resource as ⊝ Already exists rather than re-creating it.
9. Enterprise mode: confirm the target org belongs to the enterprise and the operator can create .github-private before running gh repo create. Never force-push or overwrite an existing .github-private default branch.
10. Enterprise mode: never claim to have automated the AI-controls designation or ruleset — these are UI-only, enterprise-owner actions. Hand them off with the exact navigation path and stop.

Suggested Agent Flow

Repository CI/CD onboarding

First-turn rule: the very first response to any onboarding request must be a gated handoff — surface prereq results and collect required inputs. It must NOT be a walkthrough, a full set of CLI commands, or a completion report. The agent must not narrate or execute onboarding steps until: (a) prereq check confirms ✅ READY, and (b) all five required inputs from step 2 are in hand.

1. Run /prereq-check to validate tools and auth. Surface the full results table — tool versions, Azure CLI auth status, GitHub CLI auth status, and a ✅/❌ per check. If CLI commands cannot execute in the current environment, present the required checklist items and ask the user to confirm each one passes manually (az ≥ 2.50 installed and authenticated, gh ≥ 2.0 installed and authenticated, jq ≥ 1.6, git installed). Never advance to step 2 until prereq results are confirmed — this is a hard gate.
2. Collect the required inputs. Ask for — and wait for answers to — at minimum: (1) target GitHub repository URL, (2) Azure subscription ID (or one per environment for multi-env), (3) RBAC role to grant (Contributor or Owner), (4) onboarding mode (single or multi-environment), (5) default branch (confirm main or ask if non-standard). Do not proceed to step 3 without all five.
3. Validate current Azure/GitHub auth context (subscription, tenant, GitHub org).
4. Ask for final confirmation.
5. Execute the required Azure CLI and GitHub CLI commands directly from this playbook.
6. Scaffold workflow files and copilot-instructions.md via ./scripts/scaffold-repo.sh on macOS/Linux/WSL, or pwsh ./scripts/scaffold-repo.ps1 on Windows (Step 9 in playbook). Report which files were created vs skipped.
7. (Optional) Offer to onboard the drift detector workflow by provisioning COPILOT_GITHUB_TOKEN (Step 10 in playbook). Skip if the user does not want scheduled drift detection.
8. Ask compliance framework and enforcement mode preferences (Step 11 in playbook).
9. Update copilot-instructions.md with compliance preferences — or, if the file was skipped by the scaffold step, surface the preferences in chat for manual integration.
10. Summarize outcome (including scaffolded file counts) and suggest verification commands.

Enterprise distribution

1. Confirm the target org/enterprise, ownership, and that gh is authenticated with repo-create permission.
2. Echo the plan (create <org>/.github-private, scaffold standards) and ask for final confirmation.
3. Create/clone .github-private and run scaffold-enterprise.sh / scaffold-enterprise.ps1 against the clone (Steps 2–3 of the enterprise playbook).
4. Review managed-settings.json (optionally enable ape-context@git-ape), validate the JSON, and have the user commit & push (Step 4).
5. Hand off the UI-only steps to an enterprise owner: AI-controls designation + ruleset (Steps 5–6).
6. Provide the verification commands and remind the user that each deploying repo still needs the repository CI/CD onboarding for Azure access.

Known Gotchas

GitHub Org Custom OIDC Subject Template (e.g. Azure org)

Some GitHub organizations (notably the Azure org) override the default OIDC subject claim template to use numeric ID-based subjects instead of name-based ones.

The skill auto-detects this by calling:

gh api "orgs/{org}/actions/oidc/customization/sub" --jq ".use_default"

• Returns true → standard format: repo:Azure/git-ape:pull_request
• Returns false → ID format: repository_owner_id:6844498:repository_id:1184905165:pull_request

If OIDC login fails with AADSTS700213: No matching federated identity record, the federated credential subjects don't match what GitHub is presenting. Fix by re-running onboarding (the skill will auto-detect and use the correct format), or manually updating existing credentials:

# Get repo/owner IDs
gh api repos/Azure/git-ape --jq '{repo_id: .id, owner_id: .owner.id}'

# Update each federated credential with correct subject
az ad app federated-credential update \
  --id <APP_OBJECT_ID> \
  --federated-credential-id <CRED_ID> \
  --parameters '{"subject":"repository_owner_id:<OWNER_ID>:repository_id:<REPO_ID>:pull_request"}'

Disabled Subscriptions

Azure subscriptions in a Disabled state are read-only — RBAC assignments will fail. Verify subscription state before onboarding:

az account show --subscription <SUB_ID> --query "{name:name,state:state}" -o table
# Test write access:
az group list --subscription <SUB_ID> --query "length(@)" -o tsv

Verification Commands

# Azure context
az account show --query "{name:name,id:id,tenantId:tenantId}" -o table

# GitHub auth
gh auth status

# Validate app federated credentials — check subjects match org OIDC format
az ad app federated-credential list --id <APP_OBJECT_ID> -o json | jq -r '.[] | "\(.name): \(.subject)"'

# Check GitHub org OIDC subject template (true = name-based, false = ID-based)
gh api orgs/<ORG>/actions/oidc/customization/sub --jq '.use_default'

# Get repo and owner numeric IDs (needed for ID-based subject construction)
gh api repos/<ORG>/<REPO> --jq '{repo_id: .id, owner_id: .owner.id}'

# Validate role assignments for SP (replace principal object id)
az role assignment list --assignee-object-id <SP_OBJECT_ID> --all -o table

# (Optional, drift detector) Confirm the Copilot engine credential is set
gh secret list --repo <ORG>/<REPO> | grep -q '^COPILOT_GITHUB_TOKEN' \
  && echo "✅ COPILOT_GITHUB_TOKEN set" \
  || echo "⚠️ COPILOT_GITHUB_TOKEN missing — drift workflow will fail its preflight"