Pagerunner Skill

Real Chrome automation for AI agents — authenticated sessions, PII anonymization, sealed secrets, site adapters, session checkpoints, and video recording.

Before installing or enabling this skill, verify the following: 1) Source and integrity of the 'pagerunner' binary — obtain it from a trusted, signed release (inspect the GitHub repo, release artifacts, and checksums). 2) Test anonymization in a safe environment — enable anonymize:true and confirm that get_content/evaluate results contain only tokens and that de-tokenization happens only on DOM writes as documented. 3) Assume the tool can read any page your browser can; avoid running on your 'personal' profile and instead create a dedicated, minimal agent profile with only the sites needed. 4) Review and lock down config: set recording/retention low (or disabled), set allowed_domains to limit navigation, and disable auto_record for personal profiles. 5) Limit autonomous invocation until you trust the binary: prefer user-invoked runs or run the daemon under a restricted service account / sandbox. 6) Confirm where KV results go and who can read them (avoid sending sensitive content to external systems). 7) If you cannot validate the binary and anonymization behavior, do not grant this skill access to real Chrome profiles — run in an isolated VM/container or decline installation. Additional useful checks: inspect the GitHub repo referenced in SKILL.md, review the pagerunner binary source or build from source, and audit ~/.pagerunner files after a test run.

Type: OpenClaw Skill Name: pagerunner-skill Version: 1.3.1 The Pagerunner skill bundle provides a comprehensive framework for Chrome browser automation by AI agents. While the tool possesses high-privilege capabilities, such as accessing real user profiles and session cookies, it includes extensive and well-documented security guardrails. Key safety features include local PII anonymization (tokenization/redaction), SSRF protection (blocking internal/private IPs), domain allowlisting, and a 'sealed secrets' mechanism designed to prevent credentials from ever being exposed to the LLM or audit logs. The provided scripts (e.g., debug-selector.sh, profile-perf.sh) and documentation (SECURITY.md, REFERENCE.md) are entirely consistent with the stated purpose of professional, security-conscious automation.