The Agent Skills Directory

[SAFE]: The skill instructions focus on administrative workflows for updating existing software installations. It explicitly directs the agent to identify and use the current installation's "owner" (service manager or runbook) to perform updates, preventing the creation of unauthorized backdoors or redundant services.
[SAFE]: Command execution is limited to established CLI tools (openclaw update status, health) and native OS management paths (e.g., macOS Keychain, LaunchAgents). The instructions specifically warn against inventing "headless workarounds" or bypassing native security controls.
[SAFE]: External references are directed to a pinned, public GitHub repository for the OpenClaw Foundation. The UPSTREAM.json and references/upstream-link-map.json maintain clear provenance and resolve links to specific, immutable commits, which is a security best practice for managing dependencies.
[SAFE]: The skill explicitly defines boundaries for data privacy, instructing the agent to keep private access details and state out of public artifacts and to report exact blockers privately to the user.

openclaw-update