The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes platform-specific shell commands to launch local HTML previews in the user's default browser. Evidence: Uses open on macOS, Start-Process in PowerShell on Windows, and xdg-open on Linux.- [DYNAMIC_EXECUTION]: The skill dynamically assembles and executes curl scripts to interact with the Zhihu OpenAPI. Evidence: publish-openapi.md provides templates for both POSIX and PowerShell that incorporate environment variables, authentication headers, and request bodies stored in local files.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied Markdown to generate Zhihu-compatible HTML, creating a surface for injection. Evidence: Ingestion points: The skill reads user messages to populate draft/latest-draft.md. Boundary markers: Not explicitly used in prompt templates. Capability inventory: Includes file system writes, network requests via curl, and shell command execution via open. Sanitization: Implements a strict HTML whitelist in conversion.md that strips dangerous elements such as <script>, <iframe>, and event handlers like on*.- [REMOTE_CODE_EXECUTION]: The update mechanism fetches version metadata from a remote endpoint and executes package management commands based on the response. Evidence: update.md fetches a manifest from openapi.zhihu.com and executes npx skills add {repository} after user confirmation.- [DATA_EXFILTRATION]: The skill reads sensitive credential data from a dedicated local configuration file for use in API requests. Evidence: Accesses ~/.zhihu/openapi-credentials.json to retrieve ZHIHU_OPENAPI_APP_KEY and ZHIHU_OPENAPI_APP_SECRET.

zhihu-sync