Security audit
jun-invest-option-master
Security checks for vulnerabilities and agentic risk
Overview
The skill largely matches an installer for an investment-agent workspace, but it has several coherence and safety concerns (hard-coded user paths, commands that fetch/replace code and register agents, and install scripts that will be executed) that you should review before running.
This skill appears to be an installer/updater for a local investment-agent workspace and mostly behaves like one, but proceed cautiously. - Do not run the auto-install or install scripts without review. Inspect scripts/auto-install.sh and scripts/install.sh first to see exactly what they do (file writes, deletes, permission changes, gateway restart). Run them only in a safe, non-production environment (or a throwaway VM/container) first. - The SKILL.md and agent docs hard-code the path /Users/lijunsheng/.openclaw/workspace-jun-invest-option-master; change that to a workspace path you control before running. Hard-coded user paths are a red flag—ensure the installer will not overwrite important files. - 'clawhub update ... --force' will fetch code remotely and can replace local files; validate the fetched code (or pin to a known-good commit/version) before running install scripts. - The package can register an agent (openclaw agents add) and may restart gateway services. Ensure you are comfortable with those changes and have backups/config export of your OpenClaw state. - Review included code (especially scripts and any files that would be executed) for any network exfiltration, secrets handling, or privilege escalation. The included adapters perform network calls to stooq/yfinance and may require futu-api and a local OpenD; no secrets are embedded, but broker credentials would be needed to fully use broker features. - If you lack the ability to safely audit or run the installer in isolation, prefer the authors' recommended 'jun-invest-option-master-installer' (the SKILL.md says this one is deprecated) or contact the owner for an auditable signed release/version. If possible, run the install steps manually one-by-one rather than allowing an automated agent to run them autonomously. What would change this assessment: explicit, reviewable contents of the install scripts showing only workspace-local copies (no network pulls or privileged ops), removal of hard-coded paths, and an install flow that fetches code from a verifiable signed release (or includes a pinned commit) would move this toward benign. Conversely, evidence that the install scripts execute arbitrary upstream code or attempt to change system-wide configs without explicit confirmation would raise the severity.
SkillSpector findings are pending for this release.