skills.rest
gozonerd
gozonerdCOMMUNITY
Explore all skills by this author →
GitHub

app-update-rec

Assess Claude app update risk and back up Claude state to GitHub before relaunching.

Updated Apr 11, 2026
Search Tags:#github#claude#restore#backup#data loss prevention#powershell#update risk
One-click install
npx skills add https://github.com/gozonerd/shadow-ai-assessment --skill app-update-rec-gozonerd
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: app-update-rec
Source: https://github.com/gozonerd/shadow-ai-assessment/tree/main/.claude/skills/app-update-rec
Command: npx skills add https://github.com/gozonerd/shadow-ai-assessment --skill app-update-rec-gozonerd

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Clicking "Relaunch" on a Claude Desktop or Claude Code update is risky — prior Anthropic app updates have caused real data loss, including a full user-profile deletion and a session-index wipe. This skill gives you a researched go/no-go verdict for the exact version you are about to install, plus a verified backup of your irreplaceable Claude state so any failure is reversible. ## Core Features & Use Cases - Live update-risk research: Three parallel agents scan GitHub issues, Anthropic official channels (status page, release notes, changelogs), and community sources (Reddit, HN) to produce a 🟢/🟡/🔴/⚪ verdict with a cited evidence ledger. - Tiered verified backup: Snapshots five tiers of Claude state (per-project memory dirs, ~/.claude tree, Desktop config, dirty git worktrees as bundles, loose untracked files) with robocopy, SHA256 manifests, and a push to a private GitHub repo as commit plus release asset. - Safety guardrails: Refuses to run while any Claude process is active, and requires a typed acknowledgment string before proceeding on a 🔴 RED verdict. - Use Case: Claude Desktop prompts you to relaunch for version 1.7196.0 — invoke the skill to learn whether that version has reported data-loss issues and snapshot your memory and config before deciding. ## Quick Start Ask the AI to run /app-update-rec with the version number shown in the update prompt to get a verdict and a verified backup before relaunching.

Frequently Asked Questions about app-update-rec

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I check if a Claude Desktop update is safe to install?▼

Invoke the skill with the pending version number, for example /app-update-rec 1.7196.0. It runs three parallel research agents across GitHub issues, Anthropic official channels, and community threads, then returns a GREEN, YELLOW, RED, or ABSTAIN verdict with a cited evidence ledger.

How do I back up Claude Code memory and config before an update?▼

The skill's backup phase snapshots five tiers: per-project memory dirs, the whole ~/.claude tree, Claude Desktop config paths, dirty git worktrees as bundles, and loose untracked files. It writes a SHA256 manifest and pushes the snapshot to a private GitHub repo as a commit plus release asset.

Does this skill work on macOS or Linux?▼

No, version 1 is Windows-only. Path detection relies on PowerShell, Get-AppxPackage for MSIX installs, and known Windows directories like %APPDATA%\Claude. Cross-platform support is explicitly deferred to a future version.

What happens if a Claude process is still running during backup?▼

The skill refuses to proceed. Test-AppRunning.ps1 checks for running Claude processes and exits with an error listing what must be quit first. There is no override flag, per a locked design decision.

Can I restore a Claude state snapshot from GitHub?▼

Yes. Run Restore-ClaudeState.ps1 with -SnapshotId and -FromGitHub to download the release asset from the backups repo and restore it. Restore defaults to dry-run preview; pass -Force to actually write, and memory restores verify against the manifest's SHA256 hash.

What are the limitations of the update verdict research?▼

The verdict depends entirely on live agent findings, not the model's training data. If all three research agents fail or find no signal, the verdict is ABSTAIN, meaning the install decision is left to the user with no signal either way.

Related Skills

Explore More Software Engineering Skills→
AGIBuildAGIBuild
10

dotnet-msix

Automate MSIX package creation, signing, and distribution for .NET desktop applications.

Community
Advanced
whmathews15whmathews15
1

dex-rollback

Restore the previous Dex version using Git backup tags and verify configuration.

Community
Advanced
wyre-technologywyre-technology
39

BetterStack Uptime

Manage BetterStack uptime monitors, incidents, heartbeats, and status pages.

Official
Intermediate
luiztrilhaluiztrilha

screenshot

Capture desktop, region, or application screenshots using system tools and Python scripts.

Community
Advanced