test-plugins-locally
Skill3 starsChanged 48 days agoMITView on GitHub
Set up, reset, or run a local testbed for the plugins in this repo. Use whenever you want to dogfood a plugin from this working copy (any branch) against a clean sibling sandbox project — covers marketplace registration, install/update, per-plugin smoke recipes.
- Deletes or force-pushes
In this repository6 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- doc-auditSkill
- feature-doneSkill
- progressive-disclosureSkill
- release-docsSkill
- validateSkill
The file
We split this skill into the rules it gives an agent
11 rules. About 0 are common, 4 are distinctive, and 7 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 25
Run only from inside the claude-plugins repo root (the directory containing .claude-plugin/marketplace.json).
- Very specific: line 29
Use AskUserQuestion (single-select) with these options:
- Distinctive: line 40
Run these probes and report what you find — do not modify anything.
- Distinctive: line 115
After install, tell the user: a Claude Code restart is required for newly installed plugins to load — surface this explicitly, don't assume …
- Very specific: line 155
After rsync, verify zero drift with the Status step's diff -rq block — it should print "(empty = match)" for the plugin(s) you just synced.
- Very specific: line 157
Before rm -rf, confirm with AskUserQuestion if the testbed has contents — never delete without an explicit yes, per the …
- Very specific: line 163
Ask which plugin to smoke-test (single-select).
- Very specific: line 212
Run /test-plugins-locally → Reset (rsyncs source → cache, bypassing the version-gated claude plugin update)
- Distinctive: line 224
Don't delete the testbed without confirming: If a previous test left useful state, the user wants a chance to inspect it first.
- Very specific: line 232
Every recipe instructs the user to cd + run claude in a fresh shell — never inline-execute the plugin from this session.
- Distinctive: line 233
After install or update, surface the restart Claude Code requirement explicitly.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/test-plugins-locally — Local Plugin Testbed
Spin up (or reset) a sibling sandbox project, point Claude Code at this repo's plugins via the local marketplace, and run a smoke recipe for the plugin under test.
What this skill does
The five plugins in this repo (onboard, notify, handoff, walkthrough, lens) ship through the apurvbazari-plugins local marketplace which points at this directory. Plugins are copied into ~/.claude/plugins/cache/apurvbazari-plugins/<plugin>/<version>/ on install, so source edits don't reach the running session until the cache is refreshed.
Gotcha: claude plugin update is version-gated — it compares the plugin.json version field and is a no-op when versions match (which they almost always do during in-branch dev, since you rarely bump version per edit). The Reset step in this skill therefore mirrors source → cache via rsync directly. claude plugin update is still useful when you've actually bumped the version, but it cannot be relied on for live-edit dogfooding.
~/Desktop/projects/
├── claude-plugins/ ← THIS repo (plugin source)
└── claude-plugins-testbed/ ← sibling sandbox (created by this skill)
└── <plugin>-test/ ← one fresh subdir per smoke recipeGuard
Run only from inside the claude-plugins repo root (the directory containing .claude-plugin/marketplace.json). If you can't see that file, stop and tell the user where they actually are — running this skill from anywhere else would silently set up the wrong testbed.
Step 1: Ask what to do
Use AskUserQuestion (single-select) with these options:
- Setup — first-time wiring: register marketplace if missing, install all 5 plugins, create the sibling testbed dir
- Reset (Recommended) — mirror source → plugin cache via
rsync(bypasses the version-gatedclaude plugin update), nuke the testbed contents, recreate empty subdirs - Run plugin recipe — pick a plugin and walk its smoke recipe in a fresh sandbox subdir
- Status — read-only: report marketplace + plugin + testbed state, then stop
Dispatch on the answer to the matching step below.
Step 2: Status (read-only)
Run these probes and report what you find — do not modify anything.
# Marketplace
claude plugin marketplace list 2>&1 | grep -A1 'apurvbazari-plugins' || echo "NOT REGISTERED"
# Installed plugins from this marketplace
claude plugin list 2>&1 | grep -B1 -A2 '@apurvbazari-plugins'
# Source branch + dirty state
git -C "$(pwd)" rev-parse --abbrev-ref HEAD
git -C "$(pwd)" status --short | head -10
# Testbed
TESTBED="$(cd .. && pwd)/claude-plugins-testbed"
if [ -d "$TESTBED" ]; then
echo "Testbed: $TESTBED"
ls "$TESTBED"
else
echo "Testbed not yet created at $TESTBED"
fi
# Cache freshness — for each plugin compare source vs the version-pinned cache dir
# Layout: ~/.claude/plugins/cache/apurvbazari-plugins/<plugin>/<version>/...
for p in onboard notify handoff walkthrough lens; do
if [ ! -f "$p/.claude-plugin/plugin.json" ]; then continue; fi
VER=$(grep -m1 '"version"' "$p/.claude-plugin/plugin.json" | sed -E 's/.*"version": *"([^"]+)".*/\1/')
CACHE="$HOME/.claude/plugins/cache/apurvbazari-plugins/$p/$VER"
if [ ! -d "$CACHE" ]; then
echo "$p@$VER: cache missing — Reset needed"
continue
fi
DRIFT=$(diff -rq "$p" "$CACHE" 2>&1 \
| grep -v '\.orphaned_at\|\.bak-\|\.in_use\|CHANGELOG\|tests/' \
| head -3)
if [ -n "$DRIFT" ]; then
echo "$p@$VER: cache STALE vs source —"
echo "$DRIFT"
else
echo "$p@$VER: cache matches source"
fi
doneSurface anything noteworthy: stale cache, orphan plugin entries (plugins installed from this marketplace that no longer exist in the source — suggest claude plugin uninstall <name>@apurvbazari-plugins if found), uncommitted source changes that won't be visible until reset.
Step 3: Setup
Execute in order. Each command is idempotent — re-running is safe.
REPO_ROOT="$(pwd)"
TESTBED="$(cd .. && pwd)/claude-plugins-testbed"
# 1. Register the local marketplace if missing (--scope user makes it global)
if ! claude plugin marketplace list 2>&1 | grep -q 'apurvbazari-plugins'; then
claude plugin marketplace add "$REPO_ROOT" --scope user
else
echo "marketplace apurvbazari-plugins already registered"
fi
# 2. Install all 5 plugins (skip already-installed)
for p in onboard notify handoff walkthrough lens; do
if claude plugin list 2>&1 | grep -q "^ ❯ ${p}@apurvbazari-plugins"; then
echo "$p already installed — skipping (use Reset to refresh)"
else
claude plugin install "${p}@apurvbazari-plugins"
fi
done
# 3. Create sibling testbed
mkdir -p "$TESTBED"
echo "Testbed at: $TESTBED"After install, tell the user: a Claude Code restart is required for newly installed plugins to load — surface this explicitly, don't assume they'll know.
Step 4: Reset
Three actions: refresh cache from source, clean the sandbox, recreate the sandbox shell.
REPO_ROOT="$(pwd)"
TESTBED="$(cd .. && pwd)/claude-plugins-testbed"
# 1. Mirror source → cache via rsync.
# Why not `claude plugin update`: it is version-gated, so when plugin.json
# version hasn't bumped (typical during in-branch dev) it's a no-op and
# your edits never reach the cache. rsync forces the cache to match source
# byte-for-byte (minus VCS and ephemeral metadata).
for p in onboard notify handoff walkthrough lens; do
if [ ! -f "$p/.claude-plugin/plugin.json" ]; then continue; fi
VER=$(grep -m1 '"version"' "$p/.claude-plugin/plugin.json" | sed -E 's/.*"version": *"([^"]+)".*/\1/')
CACHE="$HOME/.claude/plugins/cache/apurvbazari-plugins/$p/$VER"
if [ ! -d "$CACHE" ]; then
echo "$p@$VER: cache missing — run Setup first"
continue
fi
rsync -a --delete \
--exclude='.git' --exclude='.orphaned_at' --exclude='tests/' \
"$p/" "$CACHE/"
echo "$p@$VER: synced source → cache"
done
# 2. Wipe the testbed (ask before rm -rf if anything is in it)
if [ -d "$TESTBED" ] && [ -n "$(ls -A "$TESTBED" 2>/dev/null)" ]; then
echo "Testbed contains:"
ls "$TESTBED"
# Use AskUserQuestion to confirm deletion before proceeding
fi
rm -rf "$TESTBED"
mkdir -p "$TESTBED"
echo "Testbed reset: $TESTBED"After rsync, verify zero drift with the Status step's diff -rq block — it should print "(empty = match)" for the plugin(s) you just synced.
Before rm -rf, confirm with AskUserQuestion if the testbed has contents — never delete without an explicit yes, per the executing-actions-with-care guidance.
Tell the user: restart Claude Code so the updated plugin cache loads. Without restart, the running session still has the pre-update copies in memory.
Step 5: Per-plugin run recipes
Ask which plugin to smoke-test (single-select). Each recipe creates a fresh <plugin>-test/ subdir inside the testbed so runs don't contaminate each other.
onboard
Onboard targets an existing codebase, so the sandbox needs something to analyze. Two options — ask which:
- Empty repo path (tests stub-empty-repo mode):
cd "$SANDBOX" && git init && claude→/onboard:start - Existing project path: copy a small project (or
git clone <some-repo>) into the sandbox, thenclaude+/onboard:start
Smoke checklist:
- Codebase analysis completes
- Wizard runs to end without crashing on a single-option
AskUserQuestion(the known guard case) .claude/,CLAUDE.md,.claude/forge-state.jsonmaterialize
notify
SANDBOX="$TESTBED/notify-test"
mkdir -p "$SANDBOX"
echo "cd $SANDBOX && claude"
echo "then run: /notify:setup, then test with: /notify:check"Smoke checklist:
/notify:setupwalks the platform detection (terminal-notifier on macOS / notify-send on Linux)- macOS: a test notification appears
.claude/settings.jsonin the sandbox gets the hook entries/notify:checkreports healthy
handoff
SANDBOX="$TESTBED/handoff-test"
mkdir -p "$SANDBOX"
echo "cd $SANDBOX && claude"Smoke flow:
- In the sandbox session, do some work (touch a file or two)
- Trigger
/handoff:save— confirm via AskUserQuestion prompt - Exit the session, start a fresh one in the same dir
- SessionStart should surface the handoff; the pickup skill should auto-route
- Confirm
.claude/handoff.mdexists between save and pickup
Notes on the live-edit loop
This is the friction the skill exists to manage:
- Edit plugin source in this repo
- Run
/test-plugins-locally→ Reset (rsyncs source → cache, bypassing the version-gatedclaude plugin update) - Restart Claude Code (mandatory — running session has cached the pre-sync copy)
cd <sandbox> && claudeand invoke the plugin- Iterate
Markdown-only edits (skill content, prompts) require the same sync + restart cycle — the cache is a literal copy, not a symlink.
Anti-patterns
- Don't
cdinto the testbed from this session. This Claude session is anchored to the plugin source. To test a plugin you need a new Claude session whose working directory is the sandbox. - Don't rely on
claude plugin updatealone for live-edit dev. It is version-gated; ifplugin.json.versionis unchanged it's a no-op and your edits never reach the cache. Reset usesrsyncto bypass this. - Don't skip the cache sync. Without it, you'll be testing the previously-installed version while believing you're testing your branch. Verify zero drift via the Status step's
diff -rqblock after a sync. - Don't delete the testbed without confirming. If a previous test left useful state, the user wants a chance to inspect it first.
- Don't try to install plugins by symlinking into
~/.claude/plugins/cache/. The marketplace flow is canonical and the symlink will be silently overwritten on the nextclaude plugin updateor reinstall.
Key rules
- Skill MUST run from the
claude-pluginsrepo root — guard on.claude-plugin/marketplace.json. - Setup is idempotent; re-running never destroys state.
- Reset is the only destructive path and MUST confirm before
rm -rfon a non-empty testbed. - Every recipe instructs the user to
cd+ runclaudein a fresh shell — never inline-execute the plugin from this session. - After install or update, surface the restart Claude Code requirement explicitly.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

