Skill Cortex Pub

What to consider before installing Skill Cortex: - Understand reflex mode: If a candidate becomes a 'reflex', the skill can skip the execution-plan confirmation and proceed after showing a notification that you must actively cancel. If you run agents unattended or are unsure how notifications are surfaced, this can lead to automatic installs/execs. Ask for a configuration option that requires explicit approval for all installs. - Trust the clawhub client and sources: The skill calls 'clawhub install' and may supplement ClawHub results with GitHub search results (marked unreviewed). Make sure your clawhub CLI is official and that you trust ClawHub/GitHub packages you may allow to be installed. Consider running this in an isolated environment (container/VM) if you allow automatic installs. - Persistent file privacy: The cortex stores routing signals and structured lessons at ~/.openclaw/skill-cortex/cortex.json. The SKILL.md requires stripping PII before writing, but that filtering is performed by the LLM instructions rather than an auditable enforcement layer. Inspect the cortex.json contents occasionally and consider encrypting or restricting its permissions. - Confirm entity-filtering & auditing: If you rely on the 'entity filtering' guarantee, ask for deterministic checks (or code) that validate the filtering, or plan to review the created cortex.json to ensure no personal data or filenames leaked in practice. - Limit capabilities if needed: If you want stronger guarantees, request configuration options such as: require explicit user approval for all installs (no reflex auto-installs), disable GitHub-supplemented installs, log and prompt before any network install, and run installs in a sandbox. - Low-risk deployment: If you still want to try it, run Skill Cortex in a controlled/test environment first, verify how it surfaces install prompts (especially reflex notifications), and review the contents of ~/.openclaw/skill-cortex/cortex.json after a few runs. If you want, I can list specific configuration checks or suggested guardrail settings to request from the skill author (e.g., disable reflex promotion, require explicit consent for GitHub-sourced Skills, or provide an allowlist for installable slugs).