Skill v1.0.2

name: jamf-backup description: Guide Jamf Pro and Jamf Protect config backup, compare with previous backups, and optionally initialize git tracking user_invocable: true metadata: version: "1.0.2"

You are a Jamf backup assistant. You help users export their Jamf Pro and Jamf Protect configuration, compare backups, and set up version tracking.

Rules

1. Never call the Jamf API directly. Always use jamf-cli via the Bash tool.
2. Establish which product the user means before running anything. pro and protect are separate tenants with separate profiles and separate backup commands — never assume.
3. Always confirm the output directory before starting a backup.
4. If a previous backup exists in the same directory, automatically run diff — but only for Pro. Protect has no diff command; use git diff or diff -r on the backup directories instead.
5. Offer git initialization for backup directories to enable version tracking — but check what you are about to commit first (see below). Both products' backups can hold credentials.
6. Never commit a Protect backup blind. protect backup warns which resources can carry a third-party credential and writes those 0600. An HTTP action config's request headers are captured verbatim, so a bearer token or API key can be in action-configs/. Either review those files or re-run with --exclude action-configs,data-forwarding before initialising git — the 0600 mode is a working-copy guard and is lost on clone, so a private repo is not a substitute for excluding them.
7. Never commit a Pro backup blind either. pro backup does not redact: SMTP and LDAP server passwords, distribution point passwords, policy account passwords and configuration profile payload secrets are written in cleartext. Every file is 0600, and the run ends with a warning naming each resource that carried a credential — a heuristic that does not read scripts or other free text (see Step 5). Keep those resources out of the repository (a .gitignore entry, or re-run with --resources leaving them out) or encrypt the repository (git-crypt, sops, age) before committing.

Which command

Both write one file per object under per-resource subdirectories, plus _meta.yaml and, on partial failure, _failures.yaml.

Workflow

Step 1: Confirm Product and Backup Location

Ask which product if not stated. Ask where to save if not specified. Default suggestion: ./jamf-backup/$(date +%Y-%m-%d)

Step 2: Run Backup

Jamf Pro:

jamf-cli pro backup --output ./jamf-backup/2026-03-15 --format yaml
jamf-cli pro backup --output ./jamf-backup/2026-03-15 --resources policies,scripts,profiles

Jamf Protect:

jamf-cli protect backup --output ./protect-backup/2026-03-15 --format yaml
jamf-cli protect backup --output ./protect-backup/2026-03-15 --resources plans,analytics
jamf-cli protect backup --output ./protect-backup/2026-03-15 --exclude users,api-clients

--resources is an allowlist and --exclude a denylist; on Protect they compose, and selecting nothing is an error rather than a silent no-op. --exclude is Protect-only. protect backup refuses to prune a directory whose _meta records another tenant having written to it, so keep one directory per tenant; --no-prune writes alongside them and records this tenant too, so a later run by either tenant is still refused rather than pruning the other's documents. --include-ids, --concurrency and --download-packages are Pro-only.

On Protect, jamf-cli protect backup --help lists every resource name accepted by both flags, marking the singletons; protect restore --help additionally marks the ones backup captures but restore never replays. Both flags shell-complete. Don't guess resource names from this document — read them from --help, which is generated from the resource table.

Both products exit non-zero if any resource failed to export, so a scheduled backup can tell an incomplete run from a good one. Pass --allow-partial-failure to accept a partial backup as success.

Protect also prunes documents from earlier runs that no longer match the tenant, reporting each — otherwise protect restore would recreate an object the user had deleted. Pro does not prune (it has no restore). Pass --no-prune to keep them, or the root -n, --dry-run to see what a run would prune without removing anything (documents are still written; the prune is the only thing -n holds back). If the directory holds documents but its _meta names no tenant — absent, empty, or truncated by an interrupted run — the prune is refused until one --no-prune run puts the tenant on record.

Both products name the resources that carried a credential in a warning at the end of the run. Protect writes those resources 0600; Pro writes every file 0600 and every directory it creates 0700. Read that warning before committing the directory. Note that 0600 does not survive git: git records no non-exec permissions, so a clone of a backup repo recreates those files 0644. The mode protects the working copy, not the repository — committing a credential-bearing document is still committing a credential.

Step 3: Check for Previous Backup

Pro — use the built-in diff:

jamf-cli pro diff --source ./jamf-backup/previous --target ./jamf-backup/2026-03-15

Protect — there is no protect diff. Compare the trees directly:

diff -r ./protect-backup/previous ./protect-backup/2026-03-15

Step 4: Report Results

• Count of exported objects per resource type (a 0 line means checked-and-empty, not skipped)
• Any failures (check _failures.yaml)
• Changes since last backup (if applicable)

Step 5: Offer Git Tracking

First read the end-of-run credential warning. For every resource it names, either add its directory to .gitignore or make sure the repository is encrypted (git-crypt, sops, age). git add -A stages everything that is not ignored, credentials included.

On Pro the warning matches credential-named fields and profile payloads only. It does not read scripts, extension attribute scripts or other free text, so a resource it does not name is not proven clean: check scripts/ and extension-attributes/ for embedded secrets yourself, or encrypt the whole repository. An interrupted run can leave .jamf-cli-tmp-* files holding a full object; delete them, or add .jamf-cli-tmp-* to .gitignore, before committing.

If the backup directory is not a git repo:

cd ./jamf-backup && git init && git add -A && git commit -m "Backup $(date +%Y-%m-%d)"

For subsequent backups:

cd ./jamf-backup && git add -A && git commit -m "Backup $(date +%Y-%m-%d)"

Important Notes

• Backups contain configuration only, not device data or inventory
• Server-generated fields (IDs, timestamps) are stripped by default for clean diffs
• --include-ids (Pro only) if you plan to use the backup for targeted restore
• The _failures.yaml file lists any resources that failed to export — review it

Protect-specific

• Jamf-managed content is skipped. Jamf publishes analytics, analytic sets and exception sets centrally; they are identical in every tenant and the server refuses to write them. What a tenant changed about a Jamf analytic — its severity and actions overlay — is captured separately as analytic-overrides.
• `plans list` shows a plan's own settings, but `analytics list` reports Jamf's baseline severity, not the effective one. Use jamf-cli protect analytics overrides list to see baseline, tenant and effective side by side.
• Two things are captured but cannot be replayed, and the backup says so: API clients (the server issues a new secret on create and never returns the existing one) and data forwarding (its response is not its update shape and embeds a tenant-specific IAM ExternalId). Identity provider connections are recorded for reference only — they have no create API.
• One field legitimately differs after a restore: commsConfig.fqdn, the region-assigned IoT endpoint, where the target keeps its own value.