Workspace Guardian

The name/description (auto-healing for OpenClaw workspaces) aligns with the included script and docs: creating memory files/directories, validating openclaw.json, checking permissions and local gateway health. The actions requested are coherent with maintaining workspace stability.

SKILL.md and references instruct reading and validating /data/.openclaw/openclaw.json, creating files, checking gateway health, and (in examples) using sed and restarting the gateway. The shipped script creates files, fixes permissions, and flags config issues but does not automatically replace API keys. The SKILL.md language about 'corrigir e reiniciar gateway' is somewhat vague — the references show example commands that would modify config in-place, so an agent following these instructions could make persistent changes to local config. This is within the claimed purpose but grants the agent discretion to modify sensitive config; the instructions could be clearer about when to prompt for approval.

No install spec; the skill is instruction-only with one shell script. Nothing is downloaded or written to disk at install time beyond the included files—this is low-risk and consistent with the skill purpose.

The skill does not request environment variables or external credentials, but it reads openclaw.json which may contain provider apiKeys and baseUrls (sensitive secrets). Reading and validating those fields is reasonable for a guardian, but users should understand the skill will access local secrets stored in that config file.

always is false and the skill does not request system-wide persistent installation. It edits/creates files inside the workspace (expected for this purpose) and references restarting the local gateway, which is an appropriate action for a startup health-check script.