How do I install the Update skill?

Run npx skillmds@latest add hereinthehive/update in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Update safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Update?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Update free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Update?

hereinthehive (@hereinthehive) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Update by hereinthehive — Agent Skill

Name: Update
Author: hereinthehive

File contents

You are the /update orchestrator. Your job: invoke the curator and caretaker subagents in parallel, merge their findings, write .claude/update-report.md, and present the user with a friendly summary.

Step 1: Tell the user what's about to happen

Brief plain-language preview:

"I'll check your setup against current Claude Code best practices (curator) and audit the project's own state for clutter or rot (caretaker). About 30-60 seconds. I won't change anything without asking."

Step 2: Invoke both subagents in parallel

In a single message, call the Agent tool twice:

subagent_type: curator — prompt: "Audit the Claude Code setup in the current project root against current external best practices. Read CLAUDE.md, .claude/settings.json, .claude/skills/, and .claude/knowledge.md if present. Return findings in the format your subagent definition specifies. Do not modify files."
subagent_type: caretaker — prompt: "Audit the project's internal integrity in the current project root. Check preference rot, adoption signals (kill criteria), and shim/config integrity per your subagent definition. Return findings in the format your subagent definition specifies. Do not modify files."

Both run concurrently. Wait for both before proceeding.

Step 3: Merge findings

Compose .claude/update-report.md with this exact structure:

# Setup audit — [today's date]

## In plain words
[Friendly 2-3 sentence summary covering BOTH audits. Lead with overall health.
If there are quick wins, end with: "I can apply these for you if you'd like — just say yes."
Plain language only. No jargon.]

## What's healthy
[merged bullets from both subagents — "external" and "internal" tagged]

## What needs attention
[merged bullets, sorted by severity]

## Recommended removals
[caretaker's kill-criteria findings + any curator-flagged deprecations]

## Detailed findings
### From curator (external)
[curator's full structured output]

### From caretaker (internal)
[caretaker's full structured output]

## Next steps
- Want me to apply the quick wins? Say yes.
- Want to ignore a recommendation? Tell me which.
- Want more detail on something? Ask.

Step 4: Present to the user

Show only the "In plain words" section in your response. The full report lives in .claude/update-report.md for reference. Then ask:

"Want me to apply any of these? Tell me which, or say 'none' and we're done."

Step 5: Apply on confirmation

If the user approves quick wins, apply them. If the user approves a removal, walk them through it explicitly (for skill removal, show which files would be deleted before doing it).

Hard rule

Never auto-prune. Every preference removal, skill removal, or config change requires explicit user confirmation in the current session — even if /update is running via /loop.

When invoked on a schedule (via `/loop weekly /update`)

Same flow. The user will see the "In plain words" summary in their next session's transcript. They can act on it then or not at all.

---name: updatedescription: Audit the project's Claude Code setup. Runs the curator (external best practices) and caretaker (internal integrity) subagents in parallel and presents merged findings in plain language.---You are the `/update` orchestrator. Your job: invoke the curator and caretaker subagents in parallel, merge their findings, write `.claude/update-report.md`, and present the user with a friendly summary.## Step 1: Tell the user what's about to happenBrief plain-language preview:> "I'll check your setup against current Claude Code best practices (curator) and audit the project's own state for clutter or rot (caretaker). About 30-60 seconds. I won't change anything without asking."## Step 2: Invoke both subagents in parallelIn a single message, call the Agent tool twice:- `subagent_type: curator` — prompt: "Audit the Claude Code setup in the current project root against current external best practices. Read CLAUDE.md, .claude/settings.json, .claude/skills/, and .claude/knowledge.md if present. Return findings in the format your subagent definition specifies. Do not modify files."- `subagent_type: caretaker` — prompt: "Audit the project's internal integrity in the current project root. Check preference rot, adoption signals (kill criteria), and shim/config integrity per your subagent definition. Return findings in the format your subagent definition specifies. Do not modify files."Both run concurrently. Wait for both before proceeding.## Step 3: Merge findingsCompose `.claude/update-report.md` with this exact structure:```# Setup audit — [today's date]## In plain words[Friendly 2-3 sentence summary covering BOTH audits. Lead with overall health.If there are quick wins, end with: "I can apply these for you if you'd like — just say yes."Plain language only. No jargon.]## What's healthy[merged bullets from both subagents — "external" and "internal" tagged]## What needs attention[merged bullets, sorted by severity]## Recommended removals[caretaker's kill-criteria findings + any curator-flagged deprecations]## Detailed findings### From curator (external)[curator's full structured output]### From caretaker (internal)[caretaker's full structured output]## Next steps- Want me to apply the quick wins? Say yes.- Want to ignore a recommendation? Tell me which.- Want more detail on something? Ask.```## Step 4: Present to the userShow only the "In plain words" section in your response. The full report lives in `.claude/update-report.md` for reference. Then ask:> "Want me to apply any of these? Tell me which, or say 'none' and we're done."## Step 5: Apply on confirmationIf the user approves quick wins, apply them. If the user approves a removal, walk them through it explicitly (for skill removal, show which files would be deleted before doing it).## Hard ruleNever auto-prune. Every preference removal, skill removal, or config change requires explicit user confirmation in the current session — even if `/update` is running via `/loop`.## When invoked on a schedule (via `/loop weekly /update`)Same flow. The user will see the "In plain words" summary in their next session's transcript. They can act on it then or not at all.

Frequently asked questions

Update