disk-doctor

--- name: disk-doctor description: Diagnose and safely reclaim disk space on an OpenClaw host --- # Disk Doctor Use this when the user asks why their OpenClaw box is filling up, reports a full disk, sees the gateway misbehaving after weeks of uptime, or asks what is safe to delete. A long-running OpenClaw host fills up quietly. The agent's own tooling caches aggressively, and none of it is cleaned automatically. On a 15 GB VPS the base install plus caches can reach 80% without the user creating a single large file of their own. ## Step 1: measure before you touch anything Never delete based on a guess. Run these first and show the user the output: ```bash df -h / du -x -h -d1 / 2>/dev/null | sort -hr | head -15 du -x -h -d1 /root 2>/dev/null | sort -hr | head -15 find / -xdev -type f -size +50M -printf '%s\t%p\n' 2>/dev/null \ | sort -rn | head -20 | awk -F'\t' '{printf "%.0f MB\t%s\n", $1/1048576, $2}' ``` `du` and `df` will disagree by a few percent. That is normal: `df` excludes the filesystem's reserved blocks, `du` does not. Neither is wrong. Report the `df` number, because that is the one that runs out. ## Step 2: check nothing is running Cleaning a cache while a package install is in flight can fail that install. ```bash uptime ps -eo pcpu,pmem,etime,args --sort=-pcpu | head -12 ``` If load is high or you see `npm`, `apt`, `pip` or a build running, wait. ## Step 3: reclaim, safest first These three are safe on any host. The first two are pure caches that regenerate on demand. The third trims archived logs and keeps the recent ones. Nothing the user created is touched. ```bash npm cache clean --force # commonly the single biggest win apt-get clean # package archives, already installed journalctl --vacuum-size=64M # keeps recent logs, drops the archive ``` Measured on two independent production OpenClaw hosts: a cleanup pass took one from 78% to 67% used, reclaiming 1.6 GB, with the npm cache alone accounting for 1.4 GB. ## Step 4: the big offenders, with judgement Report these to the user and let them decide. Do not delete them unprompted. **Docker.** Docker cleanup belongs here, with the user's explicit approval. A stopped container keeps its own writable layer, and whatever was written there exists nowhere else unless it lives in a named volume or a bind mount. Removing the container removes that data for good. Show the state first: ```bash docker system df docker ps -a docker image ls docker network ls ``` Name the exact containers, images and networks you propose to remove, explain which ones hold data outside a volume, and wait for a clear yes. When the goal is only to reclaim build cache, use the narrow command and let Docker ask for its own confirmation: ```bash docker builder prune ``` **Browser engines.** `~/.cache/ms-playwright` holds a full Chromium per version, and versions accumulate: each new one lands next to the old. Measured on a live host: 389 MB and 364 MB for two Chromium builds, plus 262 MB and 254 MB for their headless shells, about 1.27 GB in total. Dropping one superseded version pair frees roughly 620 MB. Old versions are only safe to remove if no installed package pins them. Check before advising: ```bash du -sh ~/.cache/ms-playwright/* 2>/dev/null | sort -hr ``` **Multi-platform binaries.** Some npm packages vendor prebuilt binaries for every platform they support. On one measured host, a single dependency carried Windows, macOS Intel, macOS ARM, Linux ARM and Linux x64 builds: about 1.65 GB total on a machine that can only ever execute one of them. Removing the foreign-platform directories is usually safe but will be undone by the next `npm install`, and can break package integrity checks. Prefer reinstalling the package with `--omit=optional` where the package supports it. **Electron downloads.** `~/.cache/electron` keeps the downloaded installer zip after extraction, 110 MB on the measured host. That copy is dead weight. ## What to leave alone - The workspace, and anything under it. That is the user's work. - `~/.openclaw` config, credentials and session state. - Docker volumes, and any stopped container the user has not cleared for removal. - Anything you cannot explain to the user in one sentence. ## Prevention Suggest a monthly cache pass. It costs far less than a rescue at 95%. If the host is persistently above 80% after cleaning caches, the disk is genuinely too small for the workload, and cleaning is treating a symptom.

存档时间线

版本索引永久保留；内容副本只保留最新 1 份，更早版本仅留索引与哈希（存档时间线的证据链不会因此断裂）。

下载存档内容副本