skills-janitor
Audit, track usage, and compare your Claude Code skills with skills-janitor—nine focused slash commands and zero dependencies.
What is skills-janitor?
skills-janitor is a plugin for Claude Code that audits, tracks usage, and compares a user’s Claude Code skills. Its purpose is to help you keep your skills set organized by identifying duplicates, checking for errors or missing information, and summarizing which skills you actually invoke.
The plugin is organized as nine focused actions, each exposed as its own slash command with autocomplete. It also includes functionality to search GitHub for skills to install and to compare a skill against similar ones found on GitHub.
Key Features
- Nine audit actions exposed as individual skills/commands: /janitor-audit, /janitor-duplicates, /janitor-check, /janitor-fix, /janitor-cleanup, /janitor-report, /janitor-usage, /janitor-search, and /janitor-compare.
- Per-skill autocomplete: each action is its own slash command, making it easier to discover the available capabilities directly in the Claude Code UI.
- Usage tracking from conversation history: parses conversation history to show which skills you invoked versus which you never used.
- Integrity checks and cleanup workflows: includes commands to check skills for errors/missing info and to remove broken skills (destructive operations are preview-first via dry-run).
- Duplicate detection: can find duplicate skills that effectively do the same thing.
- Skill discovery and comparison on GitHub: searches GitHub for skills by keyword and compares your skill with similar ones on GitHub.
How to Use skills-janitor
- Install the plugin (one of the documented methods):
- From the Claude Code marketplace by adding
khendzel/skills-janitor, or - Clone the repository directly into
~/.claude/skills/skills-janitor.
- From the Claude Code marketplace by adding
- Run commands via slash syntax. Examples shown in the project:
- Inventory:
/janitor-audit - Usage:
/janitor-usage - Search:
/janitor-search - Compare:
/janitor-compare - Validation:
/janitor-check - Auto-fix (preview-first):
/janitor-fix
- Inventory:
- Optionally use keywords in natural language. The plugin can trigger skills based on keywords in skill descriptions (examples given include “audit my skills”, “which skills do I use?”, “search for n8n skills”, and “compare my-skill against alternatives”).
- Review dry-run outputs before changes. The project states that destructive operations use dry-run by default and it won’t delete anything without explicit confirmation.
Use Cases
- Monthly skill hygiene: generate a full report of your installed skills and identify those with issues, duplicates, or missing information using /janitor-report and /janitor-cleanup.
- Spotting unused skills: run /janitor-usage to see which skills are actively used versus never invoked, then decide which ones to remove.
- Reducing redundancy: use /janitor-duplicates to find skills that perform the same job so you can consolidate your skill set.
- Debugging skills with missing details: run /janitor-check to surface errors or missing information, then follow up with /janitor-fix (dry-run by default) to preview changes.
- Finding and comparing new skills: search GitHub for skills with /janitor-search and evaluate alternatives with /janitor-compare before installing or updating your skill library.
FAQ
-
Does skills-janitor ever delete skills automatically? The project explicitly states it will never delete anything without explicit confirmation.
-
Are destructive operations applied immediately? The repository states that destructive operations use dry-run by default.
-
Will it modify marketplace or plugin-provided skills? It states it will never modify plugin/marketplace skills.
-
What does the usage report rely on? It parses your conversation history to determine which skills you invoke and which you never use.
-
What do I need locally to run the plugin code? The requirements listed are Bash, Python 3, and curl (the repository also notes there are no pip installs and no node modules).
Alternatives
- Manual skill curation in Claude Code: keeping your skills organized by reviewing installed skills and selectively removing or updating them without automated auditing; this is simpler but lacks structured inventory, duplicate detection, and usage summaries.
- Generic GitHub search + README comparison: searching GitHub for skills and comparing repositories yourself; this can work for discovery and comparison but won’t provide the plugin’s usage tracking or built-in health checks.
- Other Claude Code skill management tools (category alternative): tools that focus on organizing, validating, or listing available skills; these may differ in whether they emphasize usage analytics, duplicate detection, or GitHub-backed discovery.
- Scripted audits using a custom CLI: building or using local scripts to scan your
~/.claude/skillsdirectory; this can offer automation but requires more setup than the plugin’s slash-command workflow.
Alternatives
Struere
Struere is an AI-native operational system that replaces spreadsheet workflows with structured software—dashboards, alerts, and automations.
Falconer
Falconer is a self-updating knowledge platform for high-speed teams to write, share, and find reliable internal documentation and code context in one place.
BookAI.chat
BookAI allows you to chat with your books using AI by simply providing the title and author.
Rectify
Rectify is an all-in-one operations platform for SaaS, combining monitoring, analytics, support, roadmaps, changelogs, and agent management—via conversation.
Yorph AI
Yorph AI is an agentic data platform combining no-code ease with code-first control and scalability for on-demand modern data work.
SaveMRR
SaveMRR scans Stripe billing data to find SaaS MRR leaks and help recover failed payments, save cancellations, and win back churn—60-sec Free Revenue Scan.