بنقرة واحدة
amp-show-changes
Preview what will change between your installed Amp and latest production without applying updates
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
القائمة
Preview what will change between your installed Amp and latest production without applying updates
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Extract and track commitments from meeting notes - who promised what to whom, with deadlines and status tracking.
Route tasks to Pi for comparison testing. Toggle Pi mode or run specific commands through Pi.
View and manage Pi-built extensions synced to Amp. Shows available tools, commands, and sync status.
Review proposed actions surfaced by the B-1 Activation Engine. Gathers recent signals, extracts candidates, ranks them, drafts artifacts, and captures the user's response per offer. Standalone — does not modify /daily-plan.
Configure AI model options - budget cloud models (save 80%+) and offline mode (use Amp without internet)
Check your AI model configuration - see what's set up, current model, credits remaining
| name | amp-show-changes |
| description | Preview what will change between your installed Amp and latest production without applying updates |
Show a read-only preview of the changes available before running /amp-update. This command helps users understand what is new, what may be removed or renamed, what customizations will be protected, and whether any breaking-change warning is available.
When to use:
/amp-update when you want a plain-English preview./amp-whats-new says an update is available and you want file-level detail.Read-only guarantee: This workflow does not merge, reset, checkout, stash, commit, or apply updates. It only reads local files and, when Git is available, runs git fetch so the remote comparison point is current.
$ARGUMENTS is optional.
Supported modes:
HEAD with the latest production main.--no-fetch: skip network fetch and compare against the currently cached remote-tracking branch.--full: include raw commit subjects and a longer changed-file summary.Check whether the current Amp install is a Git repository:
git rev-parse --is-inside-work-tree
If Git is unavailable or the command fails, continue with a limited local-only preview:
⚠️ Git preview unavailable
I could not compare your install with production because this folder is not a Git checkout or Git is not installed.
What I can still show:
- Protected customizations I can detect locally
- Manual update guidance
To enable full previews, install Amp from the Git repository instead of a ZIP download.
When Git is available, detect the production remote. Prefer upstream, fall back to origin:
if git remote | grep -qx 'upstream'; then
AMP_REMOTE='upstream'
elif git remote | grep -qx 'origin'; then
AMP_REMOTE='origin'
else
AMP_REMOTE=''
fi
If no remote exists, explain that remote comparison is unavailable and continue with the protected customizations inventory.
Unless --no-fetch is present, fetch the production branch only:
git fetch "$AMP_REMOTE" main:refs/remotes/$AMP_REMOTE/main --prune
Set the comparison target:
AMP_TARGET="$AMP_REMOTE/main"
Validate the target exists:
git rev-parse --verify "$AMP_TARGET"
If fetch fails, use the cached remote-tracking branch if available. If no target is available, report the failure and continue with local inventory only.
A production update manifest may be available. If one exists locally or in the fetched target, use it to enrich the preview. Check likely manifest paths without assuming one is present:
for path in \
System/update-manifest.json \
System/update-manifest.yaml \
System/update-manifest.yml \
.amp/update-manifest.json \
amp-update-manifest.json \
release-manifest.json; do
if [ -f "$path" ]; then
echo "local:$path"
break
fi
if git cat-file -e "$AMP_TARGET:$path" 2>/dev/null; then
echo "remote:$path"
break
fi
done
If a local manifest is found, read it from the working tree. If a remote manifest is found, read it from the target tree with git show "$AMP_TARGET:$path". Use any available fields for:
If no manifest is found, say so plainly and generate the preview from Git diff data:
No update manifest found on production main yet, so this preview is based on Git commits and changed files.
Collect the commit range and changed files:
git log --oneline --decorate --no-merges HEAD.."$AMP_TARGET"
git diff --name-status HEAD.."$AMP_TARGET"
git diff --stat HEAD.."$AMP_TARGET"
If there are no commits in HEAD..$AMP_TARGET, report:
✅ You're already aligned with production main.
No update is waiting. I still checked protected customizations below.
For normal output, summarize:
Use manifest data first when available. Otherwise infer cautiously from file paths.
Feature path patterns:
.claude/skills/<skill-name>/SKILL.mdcore/mcp/*_server.pydocs/*.md, README.md, CLAUDE.md.template, AGENTS.mdSystem/Templates/*, 06-Resources/Templates/*scripts/*, .scripts/*From git diff --name-status:
A .claude/skills/foo/SKILL.md means added skill /foo.D .claude/skills/foo/SKILL.md means removed skill /foo.R... old new under .claude/skills/ means renamed skill.Do not overstate inferred changes. Use labels:
Use manifest breaking-change fields first. If the manifest contains breaking_changes, breakingChanges, warnings, migrations, or requires_action, surface them exactly and prominently.
If there is no manifest, scan commit subjects and changed file paths for likely warning terms:
git log --format='%s' HEAD.."$AMP_TARGET" | grep -Ei 'breaking|migration|required|remove|rename|deprecat|manual' || true
git diff --name-status HEAD.."$AMP_TARGET" | grep -Ei 'migration|template|CLAUDE|AGENTS|requirements|package-lock|install|setup' || true
Output one of:
⚠️ Breaking-change warnings
- [Manifest-provided warning or cautious inferred warning]
or:
✓ No breaking-change warning found in the manifest or commit/file scan.
When warnings are inferred, say they are inferred and recommend reading the related commits before updating.
Inventory local customizations that /amp-update should preserve. Do not print private content. Report counts and names only when names are safe and useful.
Check:
# Protected user instruction block
if [ -f CLAUDE.md ] && grep -q 'USER_EXTENSIONS_START' CLAUDE.md && grep -q 'USER_EXTENSIONS_END' CLAUDE.md; then
echo 'user_extensions_block=present'
else
echo 'user_extensions_block=not_found'
fi
# Custom skills
find .claude/skills -maxdepth 1 -type d -name '*-custom' -print 2>/dev/null | sed 's#^.claude/skills/##'
# Custom MCP servers by name only
python3 - <<'PY'
import json
from pathlib import Path
path = Path('.mcp.json')
if not path.exists():
print('custom_mcp=[]')
else:
data = json.loads(path.read_text())
names = sorted(k for k in data.get('mcpServers', {}) if k.startswith('custom-'))
print('custom_mcp=' + repr(names))
PY
# Locally modified tracked files
git status --short
Report:
🛡️ Protected customizations inventory
- User extension block: present/not found
- Custom skills: N found
- Custom MCP servers: N found
- Local uncommitted changes: N files
I did not print private instruction text, MCP command values, environment values, notes, tasks, projects, people pages, or meeting notes.
If custom items are missing, avoid implying data loss. Say only that none were detected.
Use this structure:
# Amp update preview
Compared: current HEAD → <remote>/main
Remote: <remote>
Mode: manifest-enriched | git-diff only | local inventory only
## Summary
- <N> commits available
- <N> files changed: <A> added, <M> modified, <D> deleted, <R> renamed
- Biggest areas: skills, docs, core, templates, tests
## Added features
- /skill-name, from manifest or inferred from file path
## Removed features
- None found, or list items
## Renamed features
- None found, or old → new
## Breaking-change warnings
- None found, or list warnings
## Protected customizations inventory
- User extension block: present/not found
- Custom skills: count
- Custom MCP servers: count
- Local modified tracked files: count
## What this means for you
Plain-English explanation of likely impact:
- What new capability they get
- Whether any behavior may change
- Whether anything needs review before `/amp-update`
- Whether it looks safe to proceed
## Next step
- If comfortable: run `/amp-update`
- If cautious: run `/amp-update-preflight`
- If warnings appeared: read the listed commits or release notes first
Keep the final recommendation grounded in evidence. Do not claim an update is safe if breaking warnings exist, the diff could not be fetched, or there are uncommitted local changes that may need review.