Unused Agent Skills Cleanup: Deadskills vs Skillreaper Safety Boundaries

AI operators and developers use this unused agent skills comparison to audit local skill libraries, read transcript evidence, separate report-only findings from reversible cleanup, and choose a safe maintenance path without pretending that every tool can remove files automatically.

Tool Summary

Deadskills and Skillreaper both look at local evidence instead of asking you to guess which agent skills are earning their place. The important difference is the action boundary. Deadskills reports what appears active, zombie, or dead; its README does not document automatic removal. Skillreaper starts read-only, then offers reversible cleanup actions such as quarantine, mute, restore, and usage-informed routing.

That makes the choice less about which name sounds sharper and more about your operating risk. If you only want an audit report, start with Deadskills. If you want a fuller maintenance tool that can quarantine unused items and undo the change, evaluate Skillreaper after reviewing its restore path.

Best Fit Users

• Codex and Claude Code users with many installed skills: use either tool to see which skills appear in local transcript evidence instead of pruning by memory.
• Solo developers cleaning a personal setup: start with Deadskills when you want a simple report before making any filesystem changes.
• Teams maintaining shared agent environments: consider Skillreaper when you need read-only scans, reversible quarantine, restore support, and a machine-readable report.

Setup / Working Method

Deadskills is documented as a Node 18+ tool. The README shows a one-off run with npx deadskills, a global install with npm install -g deadskills, and a local build path from the GitHub repository. It auto-detects Claude Code and Codex directories when they exist, reads local transcripts, and reports skill states.

Skillreaper documents several install routes: Homebrew, npm, npx, and Go. The commands expose either reap or skillreaper depending on the install method. Its default scan is read-only. The README also documents actions such as reap prune, reap mute, reap restore --all, reap why, reap gap, and JSON or Markdown report output.

Comparison / Limits

Deadskills is the lighter choice when you want a report about skill usage and context cost. It classifies skills as active, zombie, or dead, estimates token cost, and includes a doctor command for parser health. Its source says transcripts never leave the machine and there are no runtime network calls.

Skillreaper is broader. It scans skills, agents, MCP servers, hooks, and prose files across supported platforms, then uses transcript evidence where the platform provides readable history. It can quarantine items into a reaped/ directory with a manifest, restore them, mute descriptions, keep protected items, and produce route or APM suggestions.

The limits matter. Both tools use approximate token estimates. Deadskills tells users to run deadskills doctor if the numbers look wrong. Skillreaper says platform formats can change, OpenCode evidence needs the sqlite3 CLI, and incomplete evidence should keep items in review instead of marking them for removal.

Use Cases

• First audit: run a read-only report before uninstalling or moving any skills, then inspect the items with the highest estimated cost and lowest evidence.
• Personal cleanup: use Deadskills to identify candidates, then remove or keep them manually after checking whether old project history explains the result.
• Reversible maintenance: use Skillreaper when you want quarantine and restore mechanics instead of permanent deletion.
• Platform coverage review: use Skillreaper when your setup spans Claude Code, Codex CLI, Hermes, OpenCode, Cursor, or OpenClaw, while remembering that some platforms may be inventory-only.

Troubleshooting & Optimization

• Check evidence quality first: if transcript parsing is incomplete, do not treat a low-use result as proof that a skill is useless.
• Prefer reversible steps: report, review, quarantine, then restore if needed. Avoid permanent cleanup until you understand how each tool read your local history.
• Separate cold from broken: a skill that never fires is different from one that fires and errors. Keep that distinction in review notes before pruning.
• Protect project-specific skills: a skill may look unused globally while still being valuable for a rare client, repo, or maintenance task.

Common Mistakes

• Check Deadskills before acting: treat its output as reporting and analysis, then decide manually whether a skill should stay, move, or be removed.
• Keep Skillreaper reversible: describe reap prune as quarantine with restore support, not as permanent deletion.
• Avoid savings guarantees: use the tools to estimate context cost and cleanup candidates, but do not promise a universal token-saving result.

Unused Agent Skills FAQ

• Q: What are unused agent skills?
  A: They are installed skills that local transcript evidence does not show being invoked, or that appear cold enough to deserve review before they keep adding context weight.
• Q: Should I start with Deadskills or Skillreaper?
  A: Start with Deadskills if you want a report-only audit. Try Skillreaper when you need broader platform coverage and reversible quarantine or restore workflows.
• Q: Do these tools upload my transcripts?
  A: The first-party sources describe local transcript analysis. Deadskills says transcripts never leave the machine; Skillreaper says it is local with zero telemetry and zero uploads.
• Q: Can I delete everything marked unused?
  A: No. Treat findings as evidence for review, especially when transcript history is incomplete, platform formats have changed, or a skill is useful only in rare projects.

Use this comparison to audit your local agent setup before your next cleanup pass.

Explore more practical tools in AI Tools or workflow-oriented guides in Business & Productivity.

I hope this unused agent skills guide helps you clean agent setups with evidence instead of guesswork.

Follow @bigprompt for more practical AI tool and workflow guidance.

Related Big Prompt Hub pages:

Product Photo to Ad Creative Workflow for Package-Locked Campaigns

ChatGPT Work Local Folder Guide: Desktop Access vs Cloud Uploads

Restaurant Promotional Flyer Prompt Builds Luxury Food Ads

Low Carb Dinner Party Prompt Builds Isometric Menu Boards

Luxury Culinary Poster Prompt for Exploded-View Food Grids

For the skills you retain, set installation scope and verification.

Big Prompt Hub Review

Start with Deadskills when your immediate job is to see which skills appear unused without changing files. Move to Skillreaper when you need a maintenance workflow with read-only scanning, reversible quarantine, restore support, and broader platform inventory. Delay either cleanup action if transcript evidence is incomplete.