Secure Shopper

Asynchronous shopping research + checkout using secure-autofill (1Password-backed browser filling) with results recorded to workspace artifacts.

This skill appears to do what it says (shop, use secure-autofill, save results) but has packaging and disclosure issues you should address before installing. Specifically: - The helper scripts write artifacts to a hard-coded path (/home/miles/.openclaw/workspace/...), which reveals a user-specific path and will not work correctly for other users — ask the author to make this path configurable or relative to the current agent/workspace. - The SKILL.md depends on a separate secure-autofill skill and mentions gateway environment variables (and a non-headless Chrome) but the manifest declares no required env vars or config paths; confirm what secrets or environment variables secure-autofill actually needs and whether those will be present and limited in scope. - The skill uses vault_suggest/vault_fill to access credentials via secure-autofill. Verify you trust the secure-autofill implementation and understand which credentials it will expose and under what conditions (ensure explicit user confirmation before any checkout or purchase action). - Because the skill spawns sub-agents that browse and can log in, test it in a safe/isolated environment (or with test accounts) first to confirm it respects the stated 'accept/deny' gate and doesn't place orders autonomously. If you plan to use it: require the author to (1) remove hard-coded paths or make them configurable, (2) declare required env vars/config paths in the skill manifest, (3) document exactly what secure-autofill vault operations occur, and (4) provide a reproducible way to run onboarding that does not assume a specific home directory. If those changes are not made, treat installation as higher risk.

Type: OpenClaw Skill Name: secure-shopper Version: 0.1.0 The skill bundle is designed for asynchronous shopping and checkout, leveraging a 'secure-autofill' skill for sensitive operations. The `SKILL.md` provides clear instructions for the AI agent, including explicit guardrails such as 'Never paste secrets' and mandatory user confirmation for checkout, mitigating prompt injection risks. The JavaScript utilities (`onboard.mjs`, `task_io.mjs`) perform file I/O strictly within the skill's configuration directory and a designated workspace artifact directory (`/home/miles/.openclaw/workspace/artifacts/secure_shopping/`), using `path.join` to prevent path traversal. There is no evidence of intentional malicious behavior, data exfiltration, unauthorized execution, or persistence mechanisms.