Mapick

Ferdighetsrådgiver og personvernlag bygget for OpenClaw, ikke for Claude Code

Tested · Didn't pass

Hva den gjør

En Node-based skill manager that scans installed skills, recommends missing ones, flags unused zombies, and redacts sensitive strings before other skills see them. Den utløses on requests like what skills am I missing or clean up my unused skills. Den leser only ~/.openclaw/skills, so on a Claude Code machine it reports an empty inventory.

Testrapport

In a clean throwaway HOME the very first command, status, made a live HTTPS GET to api.mapick.ai carrying the device fingerprint and returned 200 — before any consent prompt appeared, and its own SKILL.md promises that no data is sent until you pick one of three consent options. The consent dialog only shows up later, on recommend. Function is also aimed at a different runtime: with five real skills sitting in ~/.claude/skills, mapick summary reported total 0, activation_rate 0%, and clean found no zombies, because every path in the code points at ~/.openclaw/skills; grep for .claude across the scripts returns nothing. A plain ls of the skills directory beat it outright. Credit where due: every outbound call goes through one audited function with an endpoint allowlist and a local outbound.jsonl log, which is how the unconsented ping was caught.

Testet på: 2026-07-28 · Claude Code 2.x (agent harness)

Installer

git clone --depth 1 https://github.com/mapick-ai/mapick.git
cd https://github.com/mapick-ai/mapick.git
mkdir -p ~/.claude/skills
cp -r mapick ~/.claude/skills/mapick

Kommandoer og eksempelprompter

/mapickFerdighetsrådgiver og personvernlag bygget for OpenClaw, ikke for Claude Code

Skills utløses av vanlige forespørsler — ingen kommandoer å huske. Etter installasjonen aktiverer prompter som disse skillen (på engelsk):

What skills am I missing that would actually help with this project?
Clean up the Claude skills I never use and show me the zombie ones
Is this newly installed skill actually safe for me to run?

Full rapport →