check-plugins

name	check-plugins
description	Plugin health monitor that validates installed Claude Code plugins, checks for marketplace updates, and surfaces issues by severity with delta reporting. Designed for /loop — runs periodically to keep your plugins healthy. Usage: /loop 1h /check-plugins
user-invokable	true

Plugin Health Monitor

You are a plugin health monitor running as a recurring /loop iteration. Your job: scan installed plugins, validate their structure, check for marketplace updates, classify findings by severity, and only alert on changes since the last check.

State Management

State file: ~/.claude/loop-recipes/check-plugins-state.md

On Start — Read State

1. Read ~/.claude/loop-recipes/check-plugins-state.md. If it does not exist, initialize:

   ---
   status: idle
   last_check: "1970-01-01T00:00:00Z"
   previous_findings: []
   ---
   # Plugin Health Log
   

2. If status: in-progress with a locked_by field set:

   • If locked_by timestamp is less than 2 hours old: a previous iteration is still running. Output "Previous iteration still running — skipping." and stop.
   • If locked_by is older than 2 hours: treat as stale lock (previous iteration likely crashed), clear it, and proceed.

3. Set locked_by: <current_timestamp> and status: in-progress.

4. Ensure ~/.claude/loop-recipes/ directory exists (mkdir -p).

On End — Write State

After every iteration:

• Clear locked_by, set status: idle
• Update last_check timestamp
• Store current findings as previous_findings for delta comparison next time
• Append iteration summary to log section

Iteration Logic

Step 1: Discover Installed Plugins

Scan the plugin cache directory:

ls -d ~/.claude/plugins/cache/*/* 2>/dev/null

For each installed plugin directory, read its plugin.json (or .claude-plugin/plugin.json). Build an inventory:

• Plugin name
• Installed version
• Plugin path
• List of declared skills, agents, hooks (from plugin.json or directory scan)

If no plugins are installed: output "No plugins installed." Stop.

Step 2: Check for Marketplace Updates

Scan marketplace sources:

ls -d ~/.claude/plugins/marketplaces/*/ 2>/dev/null

For each marketplace, check if any installed plugin has a newer version available by comparing the installed plugin's version against the marketplace's listing. Read marketplace manifests (e.g., marketplace.json) to find available versions.

Record any plugins with updates available.

Step 3: Structural Validation

For each installed plugin, run a quick health check:

1. plugin.json validity:

   • Is it valid JSON?
   • Does it have required fields: name, version, description?
   • Are declared paths (skills, agents, hooks) valid?

2. Skill file validation (for each declared skill):

   • Does the SKILL.md file exist?
   • Does it have YAML frontmatter with name and description?
   • Is the name in frontmatter kebab-case?

3. Agent/Hook file validation:

   • Do referenced agent/hook files exist?
   • Do they have required frontmatter fields?

4. Orphan detection:

   • Scan subdirectories of the plugin's skills/agents/hooks paths. A subdirectory is a valid component if it contains a SKILL.md, AGENT.md, or HOOK.md file. Flag subdirectories that contain these files but are not referenced by plugin.json.

Step 4: Classify Findings by Severity

Severity	Criteria	Alert?
Critical	Missing plugin.json, invalid JSON, missing required fields	Yes — always
Warning	Missing skill files, outdated version, orphan components	Yes — always
Info	Available updates, minor frontmatter issues, naming suggestions	No — log only

Step 5: Delta Reporting

Compare current findings against previous_findings from state:

• New findings: Issues found now that weren't in the previous check → alert
• Resolved findings: Issues in previous check that are gone now → log as resolved
• Unchanged findings: Same issues as before → do not re-alert

Step 6: Report

If there are new critical or warning findings, output them:

## Plugin Health Check — <timestamp>

### New Issues

#### Critical
- <plugin-name>: <description of issue>

#### Warning
- <plugin-name>: <description of issue>

### Updates Available
- <plugin-name>: <installed> → <available>

### Resolved Since Last Check
- <plugin-name>: <what was fixed>

If there are only info-level findings or no new findings: output "All plugins healthy. No new issues." (brief, non-intrusive).

If there are new critical findings, suggest running a manual validation. Note: /plugin validate may not exist in all Claude Code versions — if unavailable, suggest reading the plugin's plugin.json and SKILL.md files directly to diagnose issues.

Stop Conditions

This skill is designed to run periodically (e.g., hourly). The user should stop the loop when:

• They're done with their session
• They want to investigate reported issues manually
• All plugins are healthy and no updates are expected