The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The script scripts/content_gen.py reads the sensitive global configuration file at ~/.openclaw/openclaw.json to obtain API keys and provider settings. While this facilitates integration with the user's existing environment, it grants the skill access to platform-wide credentials.\n- [INDIRECT_PROMPT_INJECTION]: The automated comment interaction feature in scripts/comments.py presents an attack surface for indirect injection.\n
Ingestion points: Live comments are fetched from the Xiaohongshu creator platform via scripts/comments.py.\n
Boundary markers: The prompt used to generate replies lacks delimiters or instructions to isolate the untrusted comment text from the system directives.\n
Capability inventory: The skill can publish/delete posts and post replies using Playwright browser automation.\n
Sanitization: No sanitization or safety filtering is applied to the retrieved comment text before it is processed by the LLM.\n- [DYNAMIC_EXECUTION]: scripts/xhs_auto.py utilizes Playwright's page.evaluate and evaluate_handle to execute JavaScript within the browser context. This is used for legitimate UI automation tasks such as session validation, clicking elements, and calling platform APIs.

xhs-publisher