XPR Creative

Provides AI tools to generate and deliver creative content including rich markdown, PDFs, CSVs, AI-generated images, videos, web media, and GitHub repositories.

Before installing, get answers to these specific questions from the skill author and take precautions: (1) Which environment variables does the skill actually require? The code uses PINATA_JWT and PINATA_GATEWAY and likely needs a GitHub token — these should be listed in the manifest and SKILL.md. (2) Understand where content is published: Pinata pins are effectively public on IPFS and create_github_repo makes public repos — do not allow the skill to upload private or sensitive data. (3) Limit token scope: if you provide a Pinata or GitHub token, restrict its permissions and use a throwaway/sandbox account when testing. (4) Ask for documentation on create_github_repo behavior (repo visibility, naming, and whether it includes credentials). (5) If you cannot validate the above, run the skill in a restricted/sandbox environment or decline to provide secrets. Finally, consider requesting the author to update skill.json and SKILL.md to explicitly declare required env vars and the privacy implications of IPFS and public repos before enabling autonomous use.

Type: OpenClaw Skill Name: xpr-creative Version: 1.0.0 The skill is classified as suspicious due to several high-risk capabilities that, while potentially part of its stated purpose, introduce significant vulnerabilities. The `store_deliverable` tool allows downloading content from arbitrary `source_url`s, and the PDF generation process downloads images from URLs embedded in markdown, posing an SSRF (Server-Side Request Forgery) and resource exhaustion risk. Additionally, the `create_github_repo` tool can create public repositories with arbitrary file content, which could be exploited for data exfiltration if a compromised agent is tricked into uploading sensitive local files (e.g., environment variables, SSH keys) to a public GitHub repository. These are vulnerabilities that could be leveraged by prompt injection or other means, rather than direct malicious intent within the skill's code.