Read, write and organise files on your VPS from PicoClaw
Gives your PicoClaw agent full file system access to read, write, move and delete files on your VPS. Designed for lightweight automation on resource-constrained servers.
Low blast radius on the evidence available, but it is still third-party code — a quick look at the SKILL.md costs nothing.
It works with files on disk. Consider which directories the agent process can actually reach.
This skill's own description refers to "file"
These permissions are inferred from this listing — the credentials it declares, any maintainer warning, and the operations its own description mentions. ClawVault has not audited this skill's source code. Treat it as a starting point for your own review, not a substitute for one.
PicoClaw does not have a skill-install command. Capabilities come from two places: skill packages, which are SKILL.md directories under workspace/skills/, and built-in tool groups configured in config.json under the tools field, which covers web, mcp, exec, cron and skills. MCP servers are managed through the mcp CLI command group.
mkdir -p workspace/skills/file-managerPicoClaw discovers skills as directories under workspace/skills/. Each one contains a SKILL.md describing the procedure, which the agent loads on demand.
$EDITOR workspace/skills/file-manager/SKILL.mdKeep it short. Sipeed's own skill-creator guidance is explicit that the context window is a shared resource: include only what the model does not already know, and make each paragraph justify its token cost.
$EDITOR config.jsonThe tools field in config.json controls the web, mcp, exec, cron and skills groups. A skill that shells out needs exec enabled; a scheduled one needs cron. Leave filter_sensitive_data on so API keys and tokens are stripped from tool output before it reaches the model.
picoclaw mcpFor capabilities that already exist as an MCP server, the mcp command group manages those entries in your config rather than writing a skill by hand.
Commands follow the official PicoClaw documentation. Placeholders such as <owner> are filled in by the search step, because registry references are publisher-scoped.
PicoClaw filters sensitive values such as API keys and tokens out of tool results before they are sent to the model, controlled by filter_sensitive_data (on by default) and filter_min_length. Because exec grants shell access, enable it only when a skill genuinely needs it.
Deploy OpenClaw, Hermes, or PicoClaw on Bluehost with the right template for the job.
Capability Evolver is the most downloaded skill on ClawHub with over 35,000 installs and it earns that position. It is a meta-skill that lets your OpenClaw agent analyse its own runtime history, identify failures or inefficiencies, and autonomously update its approach to improve performance over time. If a task fails three times in a row, Evolver detects the pattern and either creates a workaround or adjusts the agent's behaviour without you doing anything. Think of it as a personal trainer running silently in the background. Over weeks of use the improvement is measurable — faster task completion, fewer retries, more accurate outputs.
GOG is the go-to Google Workspace skill on ClawHub with over 14,000 downloads and the most complete implementation available. A single installation gives your OpenClaw agent full access to Gmail, Google Calendar, Google Drive, Google Docs, Google Sheets and Google Contacts through one consistent connection. Instead of installing and configuring five separate Google skills, GOG handles all of them through a unified OAuth flow. For anyone whose work life runs through Google Workspace, this skill transforms OpenClaw into a genuine productivity assistant.
Memory LanceDB is OpenClaw's vector-based long-term memory plugin and one of the most impactful skills you can add to any agent running 24/7 on a VPS. Unlike basic memory that stores conversation history as flat text, LanceDB uses vector embeddings to store and retrieve memories semantically — meaning your agent can recall relevant past context even when you phrase a question differently. Every task, preference, outcome and piece of information your agent encounters gets stored and becomes searchable. The longer the agent runs the more useful it becomes.