Works with the AI Agents you already use
bix-claude-skills: Claude Code Disk Cleanup
Scan ~/.claude/, see what is safe to delete, and reclaim disk space with a before/after report.
Free
bix-claude-skills: Claude Code Disk Cleanup
Example session with this skill installed
I need to free up some space. Run the cleanup tool to see what I can delete from the Claude directory.
or run the below commands from claude code.
/cleanup
Check how much disk space Claude Code is using and clean up the stale stuff.
- Read your context and instructions
- Compiled the bix-claude-
Claude Code scanned ~/.claude/ and found 5.94 GB in total, of which 5.50 GB is in directories marked safe to delete. One session is currently running, so its transcript is protected. You selected projects, debug and file-history. The cleanup removed 5.37 GB. Your skills, agents, plugins, settings and memory were not touched.
Scan output
| Directory | Size | What it contains | Safe to delete? |
|---|---|---|---|
projects/ | 3.12 GB | Per-project session transcripts (memory/, CLAUDE.md, active sessions preserved) | Yes |
file-history/ | 1.46 GB | File edit history snapshots from past sessions | Yes |
debug/ | 788.6 MB | Debug log files | Yes |
plugins/ | 412.3 MB | Installed plugins and marketplace data (deletion uninstalls them) | No (keep) |
telemetry/ | 96.4 MB | Usage telemetry cache | Yes |
shell-snapshots/ | 31.2 MB | Shell environment snapshots | Yes |
skills/ | 2.1 MB | Global user-defined skills | No (keep) |
my-notes/ | 180.0 KB | Unknown, investigate before deleting | Unknown |
| Reclaimable total | 5.50 GB | sum of all rows marked Yes | n/a |
| ~/.claude/ total | 5.94 GB | all directories above | n/a |
1 active session(s) detected (14.8 MB) and will be preserved.
Selection (checklist): projects/ (3.12 GB), file-history/ (1.46 GB), debug/ (788.6 MB).
Report
| Directory | Before | After | Saved |
|---|---|---|---|
projects/ | 3.12 GB | 38.5 MB | 3.08 GB |
file-history/ | 1.46 GB | 0 B | 1.46 GB |
debug/ | 788.6 MB | 0 B | 788.6 MB |
Total Memory Saved: 5.31 GB
~/.claude/ is now 580.7 MB.
Next steps
- Run
/cleanup -allnext time to clear every safe directory after one confirmation. - Check
my-notes/, which is not recognized, and decide yourself whether to remove it. The skill never offers to delete unknown directories. - Close other Claude Code sessions first if you want their transcripts cleaned too.
Connects securely to your tools. The creator never sees your data.
About this skill
What it does
- Scans
~/.claude/and prints a size table sorted largest first, with each directory marked Yes, No (keep) or Unknown for deletion. - Shows a Reclaimable total and a
~/.claude/total in the table footer. - Lets you pick which safe directories to clean, or use
-allfor one confirmation. - Deletes only what you approve, then prints a before/after table with Total Memory Saved.
Problems it solves
~/.claude/grows with transcripts, debug logs, file history, caches and snapshots, and it is hard to tell what is safe to remove.- Deleting by hand risks wiping skills, agents, settings or memory.
- Removing a running session's files can break that session.
How it works
- Scan:
scan.shmeasures every directory, classifies it, detects running sessions by checking each PID, and renders the table. - Select: the agent asks you which safe directories to delete (checklist), or asks one yes/no question with
-all. - Delete:
delete.shremoves the contents of the chosen directories. A hard blocklist and the classifier refuse anything user-owned. - Report: the script's own before/after table is shown as is.
All sizes, classification and deletion are done by bundled shell scripts, not by the model, so the numbers are deterministic.
Why install it
- Free and open source (Apache-2.0).
- Nothing is deleted without your confirmation.
- Protects skills, agents, commands, rules, memory, plugins, settings, credentials and IDE state.
- No jq or Python needed. Works on macOS, Linux, WSL and Git Bash.
Requirements
- Claude Code and bash.
- A
~/.claude/directory.
- Claude Code only. It reads and deletes the
~/.claude/layout. - Deletion is permanent. There is no trash, undo or dry-run mode.
- The classifier is a fixed list. Directories it does not know are labeled Unknown and are never deleted.
- Safe directories are emptied entirely, including
plans/,todos/,tasks/,backups/andfile-history/. Review the table first if you want anything in them. - Running-session detection checks the PID in
sessions/*.json. It protects transcripts and session folders inprojects/. If you selectsessions/, its metadata files are removed along with the rest of that folder. - Deletion only touches directories inside
~/.claude/. The scan also reports the size of a Claude Desktopvm_bundlesfolder, butdelete.shdoes not remove it. - Sizes use
du, so figures can differ slightly between macOS, Linux and Git Bash. Output is rounded to one decimal. - Requires bash. On Windows use Git Bash, MSYS2, Cygwin or WSL.
- It will not run on its own. Manual invocation only (
/cleanup).
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
2 uses
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 month ago
- Free with an Agensi account
Permissions
File Scopes
Deletion is the purpose of the skill, so it needs shell and file write access, but only inside `~/.claude/` and only for directories you pick. A hard blocklist protects `skills/`, `agents/`, `commands/`, `rules/`, `memory/`, `plugins/`, `ide/`, `scripts/`, `output-styles/`, and files such as `settings.json`, `credentials.json`, `CLAUDE.md`, `.claude.json` and `.mcp.json`. It does not read the contents of credential files.