amp-update

What This Command Does

For non-technical users: Updates Amp to the latest version automatically. No command line knowledge needed — just run the command and follow the prompts.

When to use:

• After /amp-whats-new shows new version available
• When you want the latest features and bug fixes

What it handles:

• Downloads updates automatically
• Protects your data (never touches your notes, tasks, projects)
• Preserves protected user blocks and user-owned MCP entries
• Resolves conflicts with a guided choice (no manual merge editor)
• Shows clear progress and confirmation

Time: 2-5 minutes

Unsure about compatibility? Run /amp-update-preflight first for environment checks, dependency scanning, and a go/no-go recommendation.

---

Process

Step 1: Quick Environment Check

Verify Git and a valid remote are available. If issues are found, suggest running /amp-update-preflight for detailed diagnostics.

git --version && git remote -v

Detect the correct remote name. Users who cloned directly have origin. Users who forked have upstream. Check in this order:

# Prefer "upstream" if it exists, fall back to "origin"
if git remote | grep -q '^upstream$'; then
  AMP_REMOTE="upstream"
else
  AMP_REMOTE="origin"
fi

Use $AMP_REMOTE for all fetch/merge commands in this process.

If Git is missing or no remote is configured, redirect: "Run /amp-update-preflight to diagnose and fix your setup."

Check for updates:

check_for_updates(force=True)

If no updates: show "✅ You're already on the latest version" and exit. If updates available: show version summary and proceed.

---

Step 2: Pre-Update Safety Check

Before changing anything, capture a protected-content inventory. This inventory is used for post-update verification and the final update summary.

python3 - <<'PY'
import json, os, pathlib, re
root = pathlib.Path('.')
protected_dirs = ['00-Inbox','01-Quarter_Goals','02-Week_Priorities','03-Tasks','04-Projects','05-Areas','06-Resources','07-Archives']
claude = root / 'CLAUDE.md'
text = claude.read_text(errors='ignore') if claude.exists() else ''
user_block = bool(re.search(r'USER_EXTENSIONS_START.*USER_EXTENSIONS_END', text, re.S))
custom_mcp = []
if (root / '.mcp.json').exists():
    data = json.loads((root / '.mcp.json').read_text())
    custom_mcp = sorted(k for k in data.get('mcpServers', {}) if k.startswith('custom-'))
custom_skills = sorted(p.name for p in (root / '.claude' / 'skills').glob('*-custom') if p.is_dir())
inventory = {
    'user_extensions': user_block,
    'custom_mcp': custom_mcp,
    'custom_skills': custom_skills,
    'user_data_dirs': [d for d in protected_dirs if (root / d).exists()],
    'user_profile': (root / 'System' / 'user-profile.yaml').exists(),
    'pillars': (root / 'System' / 'pillars.yaml').exists(),
    'env_present': (root / '.env').exists(),
}
print(json.dumps(inventory, indent=2))
PY

Treat every item in this inventory as user-owned. Do not replace it with Amp defaults during the update.

A. Check for uncommitted changes

Run: git status --porcelain

If there are changes:

💾 Saving your work...

Amp found unsaved changes in your vault.
Let me save them before updating.

Run:

git add .
git commit -m "Auto-save before Amp update to v1.3.0"

Show:

✓ Your work is saved

B. Create backup reference (safety net)

Run:

git tag backup-before-v1.3.0

This creates a snapshot user can revert to if needed.

---

Step 4: Download and Apply Updates

⬇️ Downloading updates from GitHub...

Run:

git fetch $AMP_REMOTE

If network error:

❌ Couldn't connect to GitHub

Please check your internet connection and try again.

[Retry]
[Cancel]

Optional breaking-change registry check:

If Sprint 1's manifest exists, use it in addition to the update-checker response. Do not fail if it is absent.

for manifest in System/update-manifest.json System/breaking-changes.json core/update-manifest.json core/breaking-changes.json; do
  if [ -f "$manifest" ]; then
    echo "Using update manifest: $manifest"
    python3 - <<'PY' "$manifest"
import json, sys
with open(sys.argv[1]) as f:
    data = json.load(f)
items = data.get('breaking_changes') or data.get('breakingChanges') or []
print(json.dumps(items, indent=2))
PY
    break
  fi
done

If the manifest contains breaking changes for the target version range, show them before continuing. If no manifest exists, rely on breaking_changes from check_for_updates and release notes.

If breaking_changes: true in update response or optional manifest:

⚠️ Important: This update includes major changes

[Show what's changing]

This is safe to proceed, but some folders or configs may change.
Migration will run automatically.

[Continue with update]
[Cancel — I'll read the details first]

If cancelled, show link to release notes and exit gracefully.

---

Step 6: Apply Updates

🔄 Applying updates...

A. Merge updates

Run:

git merge $AMP_REMOTE/main --no-edit

B. Handle merge outcome

Case 1: Clean merge (no conflicts)

✓ Updates applied successfully

Continue to Step 7.

---

Case 2: Merge conflicts

Check which files have conflicts:

git status | grep "both modified"

Automatic conflict resolution (protected blocks + guided choices):

Protected user blocks (preserved verbatim):

• CLAUDE.md contains a user block:
  • USER_EXTENSIONS_START ... USER_EXTENSIONS_END

Custom MCP servers (preserved by name):

• Any MCP server name starting with custom- is preserved
• Example: custom-gmail, custom-hubspot

Custom skills (preserved by name):

• Any skill folder ending with -custom is preserved
• Example: meeting-prep-custom, daily-plan-custom

When conflicts occur:

1. If file is user data (00-07, System/user-profile.yaml, System/pillars.yaml):

   • Keep user version
   • Run: git checkout --ours <file>

2. If file contains protected user block (CLAUDE.md):

   • Take upstream version
   • Re-insert preserved user block(s) verbatim
   • Validate markers still present

3. If file is .mcp.json:

   • Preserve any MCP entries named custom-*
   • Continue with Amp core updates for all other MCPs

4. If skill folder ends with -custom:

   • Preserve entirely, never modify
   • These are user's personal skills

5. If file is core Amp (skills, core MCP, scripts) and user edited it:

   • Use AskUserQuestion to resolve, instead of a merge editor

AskUserQuestion flow (generic, parameterized):

Title: Amp update conflict: {{item_name}}

Your change:
{{user_change_summary}}
Enables: {{user_use_case_summary}}

Amp update:
{{amp_change_summary}}
Enables: {{amp_use_case_summary}}

Options:
1) Keep my version (preserve my changes)
2) Use Amp version (take upstream changes)
3) Keep both (rename one)
4) Let me tell you what to do (I'll write instructions)

If AskUserQuestion is not available (non-Claude Code):

Use the tracked CLI helper instead of opening a merge editor:

.scripts/amp-merge-resolver.sh

The helper must show each conflicted file, a conflict preview, and the same four choices:

1. Keep mine: preserve local version, skip Amp changes for that file
2. Use Amp: take the update version, discard local edits for that file
3. Keep both: keep the local file and save the Amp version as <file>.amp-update
4. Instructions: collect user instructions in System/update-conflict-instructions.md and leave the conflict unresolved for follow-up

If the user types an invalid choice twice, default to "Use Amp version" for core Amp files only. For user data, System/user-profile.yaml, System/pillars.yaml, .env, custom-* MCPs, and *-custom skills, keep the user's version.

If user chooses "Keep both":

• MCP: name → name-custom
• Skill folder: name/ → name-custom/

After resolving all conflicts:

git add <file>
git commit --no-edit

Show to user:

✓ Updates applied successfully

Handled conflicts:
• Preserved your protected blocks
• Updated core Amp features
• Resolved overlapping changes with your choice

[See what changed]

---

Case 3: Merge failed (rare)

❌ Update couldn't complete automatically

This is rare, but sometimes updates need manual review.

**What happened:**
[Error message]

**Options:**
[Restore to before update] — Uses the backup we created
[Get help] — Opens GitHub issue template

If restore:

git merge --abort
git reset --hard backup-before-v1.3.0

---

Step 7: Post-Update Steps

A. Check for migration needs

If breaking_changes was true, check for migration script:

ls core/migrations/v*-to-v*.sh

If found:

🔧 Running migration...

This update requires a one-time migration to update your data structure.
This is safe and automatic.

Run:

./core/migrations/v1-to-v2.sh --auto

Show migration output.

B. Update dependencies

📦 Updating dependencies...

Run:

npm install
pip3 install -r requirements.txt

C. Sync MCP Configuration (Automatic)

Check if new MCP servers were added in the update by comparing .mcp.json.template entries against the user's live .mcp.json.

For each entry in .mcp.json.template that is NOT in the user's .mcp.json:

1. Read the entry from .mcp.json.template
2. Replace {{VAULT_PATH}} with the actual vault path
3. Add to the user's .mcp.json
4. Log: "✓ Added new MCP server: [name]"

Never remove or modify existing user MCP entries. Only add missing ones.

Example: If .mcp.json.template has amp-analytics but user's config doesn't:

"amp-analytics": {
  "type": "stdio",
  "command": "python",
  "args": ["<vault_path>/core/mcp/analytics_server.py"],
  "env": { "VAULT_PATH": "<vault_path>" }
}

Add to summary if new MCPs added: "✓ Added new MCP servers: amp-analytics"

MCP validation after sync:

Never copy a private .mcp.json from another vault. Validate only the user's live file after template entries are merged.

python3 - <<'PY'
import json, pathlib
root = pathlib.Path('.').resolve()
config_path = root / '.mcp.json'
if not config_path.exists():
    raise SystemExit('Missing .mcp.json after update')
data = json.loads(config_path.read_text())
servers = data.get('mcpServers', {})
errors = []
for name, server in servers.items():
    blob = json.dumps(server)
    if '{{VAULT_PATH}}' in blob:
        errors.append(f'{name}: template placeholder not replaced')
    env_path = (server.get('env') or {}).get('VAULT_PATH')
    args = server.get('args') or []
    if env_path and pathlib.Path(env_path).expanduser().resolve() != root:
        errors.append(f'{name}: VAULT_PATH does not match this vault')
    for arg in args:
        if isinstance(arg, str) and 'core/mcp/' in arg:
            resolved = pathlib.Path(arg).expanduser()
            if not resolved.is_absolute():
                resolved = root / resolved
            if not resolved.exists():
                errors.append(f'{name}: arg path missing: {arg}')
if errors:
    raise SystemExit('\n'.join(errors))
print('✓ MCP config valid')
PY

If validation fails, stop before dismissing the update notification and show the exact server names to fix.

D. Sync Usage Log Features (Automatic)

Merge new feature entries from the template System/usage_log.md into the user's existing System/usage_log.md.

Merge logic:

1. Read the upstream template System/usage_log.md (from the just-updated amp-core files)
2. Read the user's existing System/usage_log.md
3. For each - [ ] or - [x] line in the template:
   • Extract the feature description (text after the checkbox)
   • Search the user's file for a line containing the same feature description
   • If found: Keep the user's version (preserves their [x] state)
   • If NOT found: This is a new feature — add it to the same section in the user's file
4. Preserve ALL user state: checked boxes, consent decisions, journey metadata, dates
5. Update the feature count in Feature adoption score: X/Y (Y = new total)

Section matching: Match new entries to the correct section by the ## Section Name headers (e.g., "## Core Workflows", "## Advanced"). If a new section exists in the template but not in the user's file, add the entire section.

Never:

• Uncheck a user's checked box
• Change consent or metadata values
• Remove entries the user has

Log: "✓ Added N new features to usage_log.md" (or "✓ Usage log up to date" if nothing added)

E. Enable new background automations (Automatic)

Check for automation scripts that need installation. These run silently without prompting.

Meeting Sync (if Granola detected):

Check if Granola is installed:

ls "$HOME/Library/Application Support/Granola/cache-v3.json" 2>/dev/null

If Granola cache exists AND meeting automation not yet installed:

# Check if already installed
launchctl list | grep com.amp.meeting-intel

If not installed:

cd .scripts/meeting-intel && ./install-automation.sh 2>/dev/null

Add to summary if installed: "✓ Enabled automatic meeting sync (runs every 30 min)"

Future automations: This pattern extends to other background services. Check for the prerequisite (e.g., app installed, API key present), then run the installer silently.

---

Step 8: Verification

✓ Update complete! Now testing...

Quick smoke test:

1. Check key files exist:

   • 03-Tasks/Tasks.md
   • System/user-profile.yaml
   • .claude/skills/daily-plan/SKILL.md

2. Check MCP configuration:

   • .mcp.json exists and is valid JSON
   • No {{VAULT_PATH}} placeholders remain
   • Every env.VAULT_PATH matches this vault
   • Every local core/mcp/*.py arg exists
   • Custom MCP entries (custom-*) from the inventory still present

3. Check custom content preservation:

   • USER_EXTENSIONS_START/END markers still present if they existed before
   • *-custom skill folders from the inventory still present
   • User data folders from the inventory still present
   • System/user-profile.yaml, System/pillars.yaml, and .env presence matches the inventory

4. Try loading user profile:

   • Read System/user-profile.yaml

5. Produce an update summary:

   • Write System/update-summary.md
   • Print the same summary in chat
   • Include versions, files changed, conflicts resolved, MCPs added, custom items preserved, validation results, and rollback tag

If all pass:

✅ Update successful!

If something fails:

⚠️ Update completed but found an issue

[Details of what failed]

Your data is safe, but you may want to:
[Restore to previous version]
[Report this issue]
[Continue anyway]

---

Step 9: Summary

Write this to System/update-summary.md and display it to the user. This file is user-local runtime output, not a production template.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
✅ Amp Updated: v1.2.0 → v1.3.0
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

What's new:
• Career coach improvements
• Task deduplication fix
• Meeting intelligence enhancement

Your data:
✓ User data folders preserved
✓ USER_EXTENSIONS preserved
✓ custom-* MCP entries preserved
✓ *-custom skills preserved
✓ user-profile, pillars, and .env presence preserved

Validation:
✓ MCP config valid
✓ Smoke checks passed
✓ Summary written to System/update-summary.md

[View full changelog]
[Start using new features]

If new automations were enabled:

🤖 New automations enabled:
✓ Automatic meeting sync (runs every 30 min)

If there were conflicts:

🔍 Changes applied:
• Updated 12 core files
• Kept 5 of your customized files
• Protected all your data folders

[See detailed change list]

---

Step 9b: Check New Integrations (After Success)

After successful update, check if new integration features are available:

from core.integrations import get_post_update_integration_message, should_show_integration_prompt

if should_show_integration_prompt():
    msg = get_post_update_integration_message()
    if msg:
        print(msg)

If integrations are available but not configured:

---

## 🔌 New: Productivity Integrations

This update includes integrations for your favorite tools:

- **Notion** — Search your workspace, pull docs into meeting prep
- **Slack** — Search conversations, get context about people
- **Google** — Gmail search, email context in person pages

**Set up now?** These are optional but unlock powerful features like:
- "What did Sarah say about the Q1 budget?" → Searches Slack
- Meeting prep pulls relevant docs from Notion
- Person pages show email/Slack history

Run `/integrate-notion`, `/integrate-slack`, or `/integrate-google` to set up.

If user has integrations that could be upgraded:

---

## 🔄 Integration Upgrade Available

You have some integrations that could be upgraded to Amp recommended packages:

### Notion
- **Current:** custom-notion-mcp
- **Recommended:** @notionhq/notion-mcp-server
- **Benefits:** Official from Notion, Best maintained, Full API coverage

**Options:**
1. **Keep existing** — Your current setup works fine
2. **Upgrade** — Run `/integrate-notion` to switch to recommended

---

Step 10: Track Usage (Silent)

Update System/usage_log.md to mark Amp update as used.

Analytics (Silent):

Call track_event with event_name amp_update_completed and properties:

• from_version
• to_version

This only fires if the user has opted into analytics. No action needed if it returns "analytics_disabled".

Clear update notification:

Call dismiss_update() from the Update Checker MCP to remove the System/.update-available file. This stops the daily update reminder from appearing in future sessions.

---

Error Recovery

If Update Fails at Any Point

User always has escape hatch:

🔙 Restoring to before update...

Run:

git merge --abort 2>/dev/null || true
git reset --hard backup-before-v1.3.0
git clean -fd

✓ Restored to v1.2.0

Nothing was changed. Your Amp is exactly as it was.

[Try update again]
[Report issue]
[Cancel]

---

Migration Support (for Breaking Changes)

Auto-Migration Flag

If migration script supports --auto flag, run non-interactively:

./core/migrations/v1-to-v2.sh --auto

Migration script must:

• Accept --auto flag
• Skip confirmation prompts
• Return exit code 0 on success
• Log to System/.migration-log

Manual Migration Required

If script doesn't support --auto:

⚠️ Manual step required

This update needs you to run a migration script.

Don't worry - it's one command and takes 30 seconds.

**In Cursor's terminal (bottom panel), run:**

./core/migrations/v1-to-v2.sh

**Then come back here when it's done.**

[I've run the migration — continue]
[Show me what the migration does]
[Cancel update]

---

Alternative: ZIP Download Path

For users who can't/won't use Git, provide manual instructions:

📥 Manual Update Method

If automatic updates don't work, you can update manually:

1. **Download latest Amp:**
   https://github.com/hemory/amp/archive/refs/heads/main.zip

2. **Copy your data and custom blocks:**
   From OLD Amp folder, copy these to NEW Amp folder:
   
   ✓ System/user-profile.yaml
   ✓ System/pillars.yaml
   ✓ 00-Inbox/ (entire folder)
   ✓ 01-Quarter_Goals/ (entire folder)
   ✓ 02-Week_Priorities/ (entire folder)
   ✓ 03-Tasks/ (entire folder)
   ✓ 04-Projects/ (entire folder)
   ✓ 05-Areas/ (entire folder)
   ✓ 07-Archives/ (entire folder)
   ✓ .env (if it exists)
   ✓ Your `USER_EXTENSIONS` block from `CLAUDE.md`
   ✓ Any custom MCP entries named `custom-*` from `.mcp.json`
   ✓ Any custom skills ending with `-custom`

3. **DON'T copy:**
   ✗ .claude/skills/ (use new version)
   ✗ core/mcp/ (use new version)
   ✗ README.md (use new version)

4. **Open new folder in Cursor**

5. **Run /setup to verify**

[Download now]
[Copy step-by-step instructions to clipboard]

---

Related Commands

• /amp-update-preflight — Pre-update checks, compatibility verification, dependency scanning
• /amp-whats-new — Check what's new without updating
• /amp-rollback — Undo last update (if something went wrong)

---

Philosophy

Automatic where possible:

• Git commands run silently
• Conflicts resolved automatically
• Dependencies updated automatically
• Migrations run automatically (when safe)

Interactive where necessary:

• Breaking changes: confirm understanding
• Manual migration: clear instructions
• Errors: always offer restoration

Safe always:

• Backup created before any changes
• User data never at risk (gitignored)
• One-command rollback if issues
• Clear status at every step

No jargon:

• Don't say "merge conflict" - say "overlapping changes"
• Don't say "upstream" - say "main Amp repository"
• Don't say "git fetch" - say "downloading updates"
• Don't say "rebase" - just don't use rebase