The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill accesses sensitive authentication data from a local file path at ~/.openclaw/secrets/xiaohongshu.json. This is required for the skill's primary purpose of social media automation but involves handling highly sensitive browser cookies.
[COMMAND_EXECUTION]: The Python script publish_long_text.py uses the Playwright library to launch and control a Chromium browser instance, which is used to automate web interactions and publish posts.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted data.
Ingestion points: The title and content parameters in publish_long_text.py receive input from the AI agent, which may be derived from untrusted external sources.
Boundary markers: No technical boundary markers (e.g., XML tags or delimiters) are implemented in the code to separate instructions from data, though the SKILL.md includes natural language warnings.
Capability inventory: The skill has the capability to perform browser automation and publish public content to a social media platform.
Sanitization: The script performs basic length validation on the title but lacks sanitization or escaping for the content field before injecting it into the web editor.
[DYNAMIC_EXECUTION]: The skill uses context.add_init_script(path=STEALTH_JS_PATH) to execute a local JavaScript file (stealth.min.js) within the browser context to evade bot detection.

xiaohongshu-publish