Skills Janitor is live · free · last verified 4 Nov 2026 · viability healthy
Skills Janitor
Skills Janitor audits Claude Code and Codex skills, subagents and MCP servers to cut wasted context
The maintainer tells you when to stop using it, which is rarer than the tool itself. If you live entirely in Claude Code and only want less context bloat, /skill-doctor and /doctor now do that from inside the harness. Keep Skills Janitor for the leftovers: Codex users, anyone vetting third-party skills for prompt injection before install, and people who want broken symlinks deleted rather than reported. Expect bug fixes only.
Verdict generatedVerified 9d ago · liveness 73/100 · cite: rightaichoice.com/tools/skills-janitor
- OpenAI Codex users who want context hygiene — Claude Code's native /skill-doctor and /doctor do nothing for them
- Developers who audit third-party skills for prompt injection before installing them
- Anyone with duplicate skills or broken symlinks sprawled across user, project, codex and plugin scopes
- Terminal-comfortable Claude Code users who want a five-minute, repeatable cleanup habit
- Claude Code-only users whose only goal is trimming context — /skill-doctor and /doctor now do it natively
- GUI-only users; every command here is a slash command, a TUI or a shell script
- Projects on GitHub Copilot, Cursor or other assistants — this targets Claude Code and Codex
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Skills Janitor if you only use Claude Code and only want to trim context — /skill-doctor, /doctor and /skills now do that natively from inside the harness.
There is no paid tier and no usage meter — the real cost is the time you spend going through a large swipe deck, which the deck orders but does not eliminate.
Free and MIT-licensed, with the whole feature set — all six slash commands, the swipe TUI, the prompt-injection scan — in one $0 tier. There is nothing to compare on price against paid context-management or MCP-observability tools; the tradeoff is maintenance mode rather than a fee.
In short
Skills Janitor — Skills Janitor audits Claude Code and Codex skills, subagents and MCP servers to cut wasted context. Best for OpenAI Codex users who want context hygiene — Claude Code's native /skill-doctor and /doctor do nothing for them, Developers who audit third-party skills for prompt injection before installing them, Anyone with duplicate skills or broken symlinks sprawled across user, project, codex and plugin scopes. Free to use.
What is Skills Janitor?
Skills Janitor — Skills Janitor audits Claude Code and Codex skills, subagents and MCP servers to cut wasted context. Best for OpenAI Codex users who want context hygiene — Claude Code's native /skill-doctor and /doctor do nothing for them, Developers who audit third-party skills for prompt injection before installing them, Anyone with duplicate skills or broken symlinks sprawled across user, project, codex and plugin scopes. Free to use.
Skills Janitor is a free, MIT-licensed CLI that inventories every skill, subagent and MCP server in a Claude Code or Codex setup, then shows you what each one costs in context. It scans user scope, project scope, codex, everything pulled in through /plugin install, and connected MCP servers — surfacing duplicates, broken symlinks, unused skills and servers that are connected but never called. Install via the plugin marketplace, npx skills add, or a direct git clone; it needs only Bash, Python 3 and curl, runs offline, and collects no telemetry. The flagship command is /janitor-swipe, a Tinder-style TUI added in v1.4. Cards are sorted heaviest-and-least-used first, so the likeliest waste sits on top: left arrow deletes, right keeps, down skips, u undoes, i inspects, q quits, and user-scope deletions stage until you confirm with y. Version 1.7 added MCP servers to that same deck, ranking connected-but-never-called servers high using real session transcripts. Usage counts include skills Claude triggered on its own, not just ones you invoked. Six slash commands cover the maintenance loop. /janitor-report gives inventory, duplicate detection and broken skills; /janitor-fix --prune actually removes dead symlinks and empty directories; /janitor-value reports always-loaded description cost against on-demand bodies; /janitor-security runs the v1.6 prompt-injection heuristic scan with PASS / REVIEW / RISK verdicts; /janitor-discover searches GitHub or vets a URL before you install, including an overlap and security check. Duplicate detection uses Jaccard similarity above a 30% threshold. Since v1.8 (September 2026) the project is in maintenance mode: bug fixes and PRs only, no new features planned. Claude Code now ships /skill-doctor, /doctor, /doctor prompt-audit and /skills natively, and the maintainer says so outright. Treat this as the janitor for gaps those leave open — Codex support, pre-install security vetting, cross-scope duplicate detection and real deletion —
Key features
- Interactive swipe TUI (/janitor-swipe) with delete, keep, skip, undo and inspect keys
- Swipe deck sorted heaviest-and-least-used first so likely waste surfaces early
- MCP server triage (v1.7) across user, project, .mcp.json and plugin-bundled servers
- Cross-references MCP servers against real session usage records
- Swiping an MCP server left removes its config entry and leaves a timestamped .bak backup
- Cross-scope duplicate detection using Jaccard similarity above a 30% threshold
- Token cost reporting via /janitor-value: always-loaded descriptions vs on-demand bodies
- Usage counts from real session transcripts, including auto-triggered skills
- /janitor-report health check for inventory, duplicates and broken skills, with --brief
- /janitor-fix --prune removes broken symlinks and empty directories
- Prompt-injection heuristic scan (v1.6+) for instruction overrides, hidden comments, zero-width unicode and base64
- Pre-install security and overlap scan via /janitor-discover on a GitHub search or URL
- PASS / REVIEW / RISK verdicts on installed skills
- Zero dependencies, offline-capable, no telemetry — runs on Bash, Python 3 and curl
- Natural-language invocation ("check my skills", "which skills are wasting context?")
Behind the verdict
Where I'd reach for Skills Janitor first: OpenAI Codex. Claude Code's native /skill-doctor, /doctor and /skills do nothing for a Codex setup, and this tool scans the codex scope alongside user and project scopes in the same pass. Second: pre-install vetting. /janitor-discover checks a URL or GitHub search hit for instruction overrides, hidden HTML comments, zero-width unicode, base64 payloads and scripts that pipe the network into a shell, returning PASS / REVIEW / RISK. The maintainers calibrated it hard for low noise — on a 178-skill machine it flagged two skills, both worth reading — and a RISK verdict means read it, not it's malware. Third: duplicate sprawl. Jaccard similarity above 30% across scopes catches the skill you cloned into two projects and forgot about. The swipe deck is the hook, and it works: sorted most-likely-waste first, so most people delete through the top five or ten cards and quit. That's a feature, not a bug — a five-minute cleanup beats a full audit you never finish. User-scope deletions stage until you confirm with y, and MCP servers swiped left lose their config entry with a timestamped .bak left behind. Plugin skills can only be flagged for review, since you can't remove a single skill from a plugin. Where it bites: every command is a slash command, a TUI or a shell script, so GUI-only users are out. The swipe TUI needs real interactive stdin — run it with the ! prefix in the terminal, not through Claude Code's Bash tool. And there's no life after v1.8. No new features, no new coverage for Copilot, Cursor or other assistants, and no guarantee the Claude Code scans keep pace as the native commands evolve. Closest alternative: Claude Code's own built-ins, and for Claude Code-only users the honest answer is to use them. They see usage from
Models under the hood
as of 2026-11-08
Limitations
- Skills Janitor is a terminal-only CLI: a swipe TUI plus slash commands, with no GUI and no cloud dashboard for centralized team management.
- Claude Code-only users who just want to trim context should start with the native /skill-doctor, /doctor, /doctor prompt-audit and /skills commands, which see usage from inside the harness.
- The tool targets Claude Code and OpenAI Codex; other assistants are out of scope.
- As of v1.8 (September 2026) the project is in maintenance mode: bug fixes and PRs are still welcome, but no new features are planned.
as of 2026-10-11
Researching Skills Janitor? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is Skills Janitor still active?
Skills Janitor is live · free · last verified 4 Nov 2026 · viability healthy.
Viability score
How well maintained and how widely used is Skills Janitor? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: November 2026
How we score →What's new in Skills Janitor
Checked 5 days agoAcross the latest 1 update: 1 changelog entry.
What do users say about Skills Janitor?
What people actually say about Skills Janitor — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
ResearchedDiscussed by users across 3 sources (YouTube, Product Hunt, Lemmy) · researched Oct 20, 2026.
Weighted by the 35 posts each of 3 sources contributed.
- +Usage stats drawn from real session transcripts, including auto-triggered skills
- +Dedup uses Jaccard similarity at a 30% threshold to catch overlapping skills
- +Token cost reporting separates always-loaded descriptions from on-demand bodies
- +MCP triage ranks connected-but-never-called servers using actual transcript records
- +Zero dependencies, offline-capable and no telemetry — praised by privacy-minded commenters
- −Project is in maintenance mode; no active feature development expected
- −Claude Code's native doctor commands now cover the main trim-context use case
- −Plugin skills and bundled MCP servers are flagged but not individually deletable
- −Little independent evidence outside Product Hunt; Reddit, HN and GitHub returned nothing
- −Early versions silently ignored the plugins tree, producing misleading audits
- • Zero monetary cost, but time spent learning six slash commands and the TUI keybindings
- • Maintenance-mode status means latent future cost if the tool stalls behind native features
How we verified this
Every listed fact about Skills Janitor — pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 4 Nov 2026. Analysis is drafted with AI models and checked against our editorial standards before publishing; see how we verify.
Verification history
We have re-verified Skills Janitor 12 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed pricing page
- — changed our verdict, pricing tiers, description, summary
- — changed pricing page
- — changed our verdict, description, pricing tiers, summary
- — changed pricing page
- — changed summary, pricing tiers, features, description
Showing the 6 most recent of 12 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does Skills Janitor cost in 2026?
Skills Janitor: Free to use. Pricing last verified 4 Nov 2026.
Plans compared
Every published Skills Janitor tier, its price as the vendor publishes it, who it fits, and what it adds over the tier below. The 12-month cost projection follows.
| Plan | Price | Billing | Who it's for |
|---|---|---|---|
| Free | Free | — | Individual developers and terminal-comfortable teams using Claude Code or OpenAI Codex who need context hygiene, duplicate detection or a prompt-injection scan.Adds: Starting tier — MIT-licensed with everything included: all six slash commands, the swipe TUI, MCP triage, duplicate detection and the security scan, with no telemetry and no paid tier above it. |
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where Skills Janitor's pricing actually pencils out — and where peers do it cheaper.
Free and MIT-licensed, with the whole feature set — all six slash commands, the swipe TUI, the prompt-injection scan — in one $0 tier. There is nothing to compare on price against paid context-management or MCP-observability tools; the tradeoff is maintenance mode rather than a fee.
Has Skills Janitor's pricing changed?
No plan price for Skills Janitor has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick Skills Janitor?
Pick Skills Janitor if you are OpenAI Codex users who want context hygiene — Claude Code's native /skill-doctor and /doctor do nothing for them, Developers who audit third-party skills for prompt injection before installing them, Anyone with duplicate skills or broken symlinks sprawled across user, project, codex and plugin scopes, Terminal-comfortable Claude Code users who want a five-minute, repeatable cleanup habit.
Look elsewhere if you are Claude Code-only users whose only goal is trimming context — /skill-doctor and /doctor now do it natively, GUI-only users; every command here is a slash command, a TUI or a shell script, Projects on GitHub Copilot, Cursor or other assistants — this targets Claude Code and Codex, Teams needing to delete individual plugin skills; the janitor only flags those for plugin review.
Skip it if Skip Skills Janitor if you only use Claude Code and only want to trim context — /skill-doctor, /doctor and /skills now do that natively from inside the harness.
GeneratedReal-world workflow fit
Concrete scenarios for the personas Skills Janitor actually fits — and what changes day-one when you adopt it.
Run /janitor-value to see which skill descriptions load on every request, then /janitor-swipe and work the deck heaviest-and-least-used first, pressing left on the first five cards and confirming with y.
Outcome: A smaller always-loaded skill set and a visible drop in per-request context, done in a single sitting.
Run /janitor-discover with a URL or search GitHub, let the pre-install security scan check for instruction-override phrasing, hidden comments, zero-width unicode and base64 payloads, and review overlap against what is already installed.
Outcome: A PASS/REVIEW/RISK verdict before the skill ever touches your config.
Run /janitor-report --brief for the inventory, /janitor-fix --prune to clear broken symlinks and empty directories, then /janitor-swipe to triage remaining dead weight.
Outcome: A clean Codex skill tree that Claude Code's native commands cannot reach.
Use cases
- Audit your Claude Code skill collection to find unused or duplicate entries across user, project, codex and plugin scopes
- Reduce context usage by deleting heavy, rarely-used skills that load on every request
- Scan installed skills for prompt injection, hidden instructions or dangerous script patterns before trusting them
- See which skills and subagents consume the most tokens per invocation versus sitting unused
- Pre-check a skill URL for security and overlap issues before installing it
- Fix broken skill symlinks and empty directories automatically with --prune
- Find MCP servers connected but silent for eight weeks and remove their tool-schema overhead
Setup time & first value
How long it actually takes to get something useful out of Skills Janitor — broken out by persona, not the marketing-page minute.
Clone the repo and it runs on Bash, Python 3 and curl with zero other dependencies — expect a few minutes to your first /janitor-report. The swipe deck is usable immediately; expect roughly five minutes for a first pass through a moderate collection, longer if you have dozens of skills.
Switching to or from Skills Janitor
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From manual `ls`-and-delete cleanup: run /janitor-report --brief for an inventory across user, project, codex and plugin scopes, then /janitor-fix --prune to clear broken symlinks and empty directories.
- →From Claude Code's native /skill-doctor and /doctor: keep using them for native context trimming, and add /janitor-security and /janitor-discover for pre-install and prompt-injection checks they do not cover.
- ↗To Claude Code native /skill-doctor, /doctor, /doctor prompt-audit and /skills: use these for context trimming if you are Claude Code-only, and keep Skills Janitor only for security scanning and Codex support.
- ↗To a context-management or MCP-observability platform: those cover the same inventory ground with dashboards and team management, but you lose the free MIT license and the actual symlink deletion.
Integrations
Resources & guides
Tutorials & Learning
YouTube returned 6 videos for “Skills Janitor”, and we withheld 6: 6 did not mention Skills Janitor. We are showing none, because we could not prove any of them are about Skills Janitor.
Official links
What are the best Skills Janitor alternatives?
The closest alternatives to Skills Janitor we list are ADE, Codecompanion.Nvim and fanbox. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| ADE | Free | Free | — |
| Codecompanion.Nvim | Free | Free | — |
| fanbox | Free | Free | — |
Alternatives to Skills Janitor
View all
ADE
ADE runs Claude Code, Codex, Cursor and other AI coding agents side by side in one synced agentic development environment
Codecompanion.Nvim
CodeCompanion.nvim brings LLM chat, inline edits and CLI agents like Claude Code and Codex into Neovim buffers.
Tools that pair well with Skills Janitor
Common stack mates teams adopt alongside Skills Janitor, with the specific reason each pairing earns its keep.
ADE
ADE runs Claude Code, Codex, Cursor and other AI coding agents side by side in one synced agentic development environment
Codecompanion.Nvim
CodeCompanion.nvim brings LLM chat, inline edits and CLI agents like Claude Code and Codex into Neovim buffers.
fanbox
FanBox is a free macOS terminal with a live diff viewer showing every file your Claude Code or Codex agent writes.
Skills Janitor vs its closest rival
Skills Janitor's closest head-to-head is with Spider Cloud.
Our Skills Janitor vs Spider Cloud verdict written Jul 3, 2026 is more than 90 days old, so it is being re-reviewed against current data and is not shown.
Head-to-head comparisons
Skills Janitor vs Spider Cloud
Verdict being re-reviewed against current pricing and data.
Skills Janitor vs Temporal Ai
Don't shortlist these against each other — they solve different problems. Buy Temporal if you need a runtime that keeps AI agent and workflow state alive across crashes, retries, and abandoned sessions, with Saga compensation and durable timers. Choose Skills Janitor if your pain is a sprawl of Claude Code and Codex skills, duplicates, and MCP servers eating your context — and note it went into maintenance mode in v1.8 now that Claude Code ships /skill-doctor and /doctor natively.
GeneratedSkills Janitor vs Voyage Ai
Verdict being re-reviewed against current pricing and data.
Frequently asked questions
Categories
Best-of guides
For makers of Skills Janitor
- Get a Verified Listing for Skills Janitor ($39): a written decision within 1 working day.
- Free listing for another tool: Free reviews are decided in order, within 3 weeks of the review starting.
Used Skills Janitor? Help shape our editorial sentiment research.