How to use the LifeOS Upgrade Scout agent skill

What is an agent skill, and what does Upgrade do?

An agent skill is a folder of instructions and optional scripts that an AI agent loads when a request matches the skill's purpose. The skill's SKILL.md file describes when it applies and how to work.

Upgrade improves a LifeOS setup in two directions. External signal comes from Anthropic's blogs, changelogs, GitHub releases and docs, plus optional YouTube channels and GitHub trending repos. Internal signal comes from your system's reflection log, which records what has gone wrong or could be better. Each candidate idea is filtered against your current files and against rejected decisions, so work already done or rejected is not recommended again.

• Checks Anthropic sources with Tools/Anthropic.ts and records state so the next run shows only changes.
• Ranks remaining ideas with cited sources and names the LifeOS file or component each would improve.
• Can keep model references current, write a sourced AI news digest, and mine reflection logs for weaknesses.
• Does not apply changes automatically. Its output needs human review.

Install the Upgrade skill from the ZIP

Download the Upgrade ZIP from this page and unzip it. Place the Upgrade folder in one of two locations, depending on whether you want it available everywhere or only in one project.

For all your projects, put it in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder. Start a new agent session afterwards so the skill is picked up.

If you use the Claude apps instead of Claude Code, you can upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/Upgrade/
• Project only: <your-project>/.claude/skills/Upgrade/
• Claude apps: Settings → Capabilities → Skills

Set up requirements before the first run

The skill needs a few tools. The Anthropic sweep runs through Bun, a TypeScript runtime, and the model check is also a Bun script. Network access to Anthropic sites and GitHub is required.

The GitHub lanes use the GitHub CLI (gh), and a GITHUB_TOKEN is optional for them. The YouTube lane needs yt-dlp and is optional. The skill also expects a working LifeOS installation, because it calls paths such as GetTranscript.ts and Paths.

For first-time configuration, the skill's Setup.md covers optional GITHUB_TOKEN, YouTube channels, user sources and the programs those lanes call.

• Required: Bun, network access to Anthropic and GitHub, and a LifeOS installation
• Optional: GitHub CLI (gh) with GITHUB_TOKEN for GitHub trending and repo lanes
• Optional: yt-dlp for the YouTube lane

Trigger the skill with natural requests

You do not need a special command. Ask in plain language and the agent matches your request to an intent. A bare request such as "upgrade" or "check for upgrades" runs the full Upgrade sweep, which includes reflection mining.

Other phrasings route to narrower jobs. Asking about models runs the model freshness check. Asking for AI news runs the news digest. The skill's routing table lists the trigger phrases for each intent.

• "Check Anthropic" or "any updates": tiered recommendations from the Anthropic sweep
• "Mine reflections" or "what have we learned": patterns in your reflection log
• "Are we on the latest models" or "model freshness": checks model IDs in your configuration
• "What's new in AI" or "AI news": a deduplicated, sourced digest
• "Competitor tracking" or "what are OpenAI and Google doing": competitor report
• "Algorithm upgrade": section-targeted proposals for the LifeOS Algorithm

Run the model check

The model check reports whether Claude model IDs in your configuration are outdated. It runs as a Bun script with a --check flag, and it exits nonzero when a Claude ID is pinned outside the registry, when agent frontmatter skips tier aliases, or when a non-Claude versioned ID sits in executable code.

Ask for a model freshness review in plain words, or run the check directly from your LifeOS folder if you prefer a command line.

Understand the output and what to do with it

Recommendations come with a Prior Status for each item. Implemented items are listed under Skipped Content with their evidence. Rejected ideas return only when the skill names a reason the context has changed.

Treat every recommendation as a proposal. The skill's own process requires a manual review that confirms each cited technique is new, applicable, and not already adopted or rejected. Sources that fail or time out are named in the report, so you can see what was not covered.

• Check that each recommendation has a supporting source.
• Check that it does not duplicate current state.
• Check that it maps to a concrete improvement in a named file.

Safety notes: scripts and network access

Upgrade runs scripts and uses the network. Its Anthropic tool fetches Anthropic sites and GitHub, and the optional lanes call yt-dlp and gh. Review the scripts in the skill folder before you run them, and know which sources are configured in your CUSTOMIZATIONS folder.

Keep GITHUB_TOKEN out of shared files. The skill does not apply changes to your system on its own, so any edit it suggests stays under your control.

• Network use is required for the Anthropic, GitHub and YouTube lanes.
• Optional lanes are skipped and noted if their config is absent.
• Security news is not covered here. That belongs to a separate SecurityUpdates skill.