auto-update（自动更新）

When to Use

Use when the user wants OpenClaw and installed skills to stay updated automatically. This skill sets up a real openclaw cron add job, keeps a small control folder in ~/auto-update/, remembers which skills should auto-update, backs up important files first, reviews migration risk before skill changes, and summarizes what changed after every run.

Architecture

State lives in ~/auto-update/. If ~/auto-update/ does not exist, run setup.md. See memory-template.md for structure.

~/auto-update/
├── memory.md        # global defaults, activation, and summary preferences
├── openclaw.md      # OpenClaw update mode, channel, backup scope, feature-review prefs
├── skills.md        # per-skill policy, installed version, backup, migration state
├── schedule.md      # approved timing, timezone, and scheduler owner
├── backups.md       # latest OpenClaw and skill backup inventory
├── migrations.md    # pending migration checks and user decisions
└── run-log.md       # recent runs, versions, and outcomes

Quick Reference

Quick Start

The default model is simple:

• create one OpenClaw cron job
• let that cron job read ~/auto-update/*.md
• update OpenClaw and only the allowed skills
• back up first, then summarize

Visible commands the user should recognize:

openclaw update status --json
openclaw update --json
clawhub update --all --dry-run
clawhub update --all

Example daily job:

openclaw cron add \
  --name "Auto-Update" \
  --cron "0 4 * * *" \
  --tz "Europe/Madrid" \
  --session isolated \
  --wake now \
  --announce \
  --message "Run the auto-update routine. Before changing anything, read ~/auto-update/memory.md, ~/auto-update/openclaw.md, ~/auto-update/skills.md, and ~/auto-update/migrations.md. Then: 1) inspect OpenClaw update status and apply OpenClaw only if openclaw.md says mode:auto 2) inspect skill updates 3) back up the approved OpenClaw files and each skill that is allowed to change 4) skip any skill marked no, pending, or ask-first 5) apply only the allowed updates 6) verify obvious health 7) write backups.md and run-log.md 8) report updated, unchanged, skipped, and failed items."

Safer variant:

openclaw cron add \
  --name "Auto-Update (Notify First)" \
  --cron "0 4 * * *" \
  --tz "Europe/Madrid" \
  --session isolated \
  --wake now \
  --announce \
  --message "Run the auto-update review. Read ~/auto-update/memory.md, ~/auto-update/openclaw.md, ~/auto-update/skills.md, and ~/auto-update/migrations.md. Inspect OpenClaw updates and run clawhub update --all --dry-run. Do not apply changes for any item in notify, no, pending, or ask-first mode. Report what would change, what is blocked, and which backups would be created."

How the Run Decides What to Do

Each cron run follows the same contract:

1. read ~/auto-update/memory.md
2. read ~/auto-update/openclaw.md
3. read ~/auto-update/skills.md
4. read ~/auto-update/migrations.md
5. inspect openclaw update status --json
6. inspect clawhub update --all --dry-run
7. back up allowed targets
8. apply openclaw update --json only if core mode is auto
9. apply skill updates only for allowed skills
10. summarize updated, unchanged, skipped, and failed items

Starter Modes

If the user says "just handle it," default to Instant daily with migration questions still enabled.

Core Rules

1. Auto-Update Means Real Scheduled Updates

• The core promise is actual OpenClaw and skill updates, not only policy notes.
• The default mechanism is an OpenClaw cron job created with openclaw cron add.
• That cron job must read the control files in ~/auto-update/ before deciding what to update.
• The same scheduled flow checks, backs up, updates, verifies, and reports for both OpenClaw and skills.
• If the user approves a daily schedule, create or update the exact scheduler entry that will run daily. Do not leave the cadence only as a note in schedule.md.

2. Learn a Default for New Skills

• Ask once whether new skills should default to all-in or all-out for auto-update.
• On every new skill install, ask two things: do they want a quick explanation of the skill, and should that skill auto-update or stay manual.
• Record the answer in skills.md so later sessions do not guess.

3. Back Up Before Changing OpenClaw or Skills

• Before OpenClaw updates, snapshot the tailored files and config the user cares about most.
• Before each skill update, save the currently installed skill folder and installed version reference.
• Log every backup in backups.md and reference it in the post-run summary.

4. Review Migration Risk Before Skill Updates

• Compare the currently installed skill state with the new version before overwriting it.
• Flag path, folder, AGENTS, TOOLS, SOUL, setup, or state-storage changes in migrations.md.
• If migration is unclear or stateful files may move, ask before applying or before first use of the new version.

5. Respect the Actual OpenClaw Update Path

• Use the documented OpenClaw path: the cron job should inspect openclaw update status --json and, if approved, run openclaw update --json from the scheduled turn.
• auto, notify, and manual live in openclaw.md; the cron message must respect them every run.
• After OpenClaw updates, run the boring checks: doctor, restart when needed, and health verification.
• If the user wants core-only automation, the cron job should skip skills explicitly instead of removing the shared control flow.

6. Turn Release Notes into Useful Suggestions

• After OpenClaw updates, summarize what changed in plain language.
• Offer an optional follow-up review that maps new features or changes to the user's actual workflow.
• Never apply workflow changes automatically just because a release note sounds promising.

7. Keep the User in Control

• Never auto-migrate state, move folders, delete backups, or rewrite workspace behavior files without approval.
• Never silently add scheduler entries or workspace reminder snippets; show the exact proposed lines first.
• Never modify this skill's own SKILL.md.
• Heartbeat is never the primary mechanism for exact daily updates. Use heartbeat only for follow-up: install-time reminders, migration reminders, failed-run review, or post-update suggestions.

Common Traps

Scope

This skill ONLY:

• configures real OpenClaw and skill update flows
• keeps local defaults and per-skill decisions in ~/auto-update/
• proposes optional install-time reminder integration for new skills
• creates backups, migration notes, and run summaries before and after updates

This skill NEVER:

• auto-migrates user state or folder structures without approval
• forces all skills into auto-update when the user chose all-out
• edits AGENTS, cron, launchd, Task Scheduler, or ~/.openclaw/openclaw.json without a visible plan or standing approval
• stores secrets in local memory files
• modifies its own skill files

Data Storage

Local state lives in ~/auto-update/:

• memory.md for durable defaults and activation notes
• openclaw.md for core updater mode, channel, backup scope, and feature review preferences
• skills.md for per-skill auto-update policy and installed version history
• schedule.md for timezone, cadence, and scheduler ownership
• backups.md for backup paths and retention notes
• migrations.md for pending migration checks and decisions
• run-log.md for compact run history and outcomes

External Endpoints

No other data is sent externally.

Security & Privacy

• This skill stores local policy and logs in ~/auto-update/.
• It may read .clawhub/lock.json, ~/.openclaw/openclaw.json, and workspace behavior files when needed for approved update work.
• It backs up files before updates, but never stores secrets in its own local ledgers.
• Scheduler changes, workspace integration, OpenClaw config edits, and risky migrations require approval unless the user has already approved that exact class of action.
• It never modifies its own SKILL.md.

Trust

By using this skill, update traffic may reach OpenClaw update sources, ClawHub, npm, or the git remote chosen by the user. Only install if you trust those services with update checks and package downloads.

Related Skills

Install with clawhub install <slug> if user confirms:

• skill-update - Review risky skill diffs, migrations, and rollback choices in more depth
• backups - Strengthen backup and restore practices beyond the default updater snapshots
• heartbeat - Pair exact-time update jobs with adaptive follow-up checks
• self-improving - Learn recurring update preferences, failure patterns, and workflow opportunities

Feedback

• If useful: clawhub star auto-update
• Stay updated: clawhub sync