The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface during its typography analysis workflow. \n * Ingestion points: 'scripts/fetch-article.mjs' fetches HTML content from external WeChat URLs (#js_content). \n * Boundary markers: There are no explicit instructions or delimiters for the agent to ignore embedded instructions in the fetched HTML. \n * Capability inventory: The skill possesses network capabilities (fetch), local command execution, and the ability to publish to external WeChat accounts. \n * Sanitization: This risk is significantly mitigated by 'scripts/validate-theme.mjs', which performs a strict safety scan on generated theme objects to block scripts, styles, event handlers, and image-src injections before they are used.\n- [CREDENTIALS_UNSAFE]: 'scripts/account-verify.mjs' accesses sensitive local file paths to retrieve service keys from '~~/.doubaoya/key' and '~~/.doubaoya/env'. It also retrieves credentials from the macOS Keychain. These actions are documented and essential for authenticating with the doubaoya.com platform.\n- [EXTERNAL_DOWNLOADS]: The skill performs network operations to 'doubaoya.com' for account verification, image generation, and publishing, and to 'mp.weixin.qq.com' for fetching reference articles. These domains are either the vendor's primary infrastructure or well-known services related to the skill's purpose.\n- [COMMAND_EXECUTION]: Legitimate use of system utilities is observed across several scripts. 'scripts/account-verify.mjs' uses 'security' to interface with the Keychain, 'scripts/design-studio.mjs' uses 'open' to provide a local GUI, and 'scripts/preprocess-and-publish.mjs' uses 'sips' for image compression.

wechat-article-pipeline