The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation and setup scripts suggest installing the Bun runtime by piping a remote script to the shell.
Evidence: curl -fsSL https://bun.sh/install | bash mentioned in SKILL.md and scripts/check-permissions.ts for environment setup.
[DYNAMIC_EXECUTION]: The skill dynamically generates helper scripts for system operations and executes them at runtime.
Evidence: scripts/copy-to-clipboard.ts and scripts/check-permissions.ts write Swift code to temporary files and execute them using the swift command to manage clipboard data (images/HTML) on macOS.
Evidence: scripts/copy-to-clipboard.ts executes dynamically constructed PowerShell commands on Windows for clipboard interaction.
[COMMAND_EXECUTION]: The skill relies extensively on executing shell commands and external processes to perform its core functions.
Evidence: Frequent use of spawnSync, execSync, and spawn to run npx, bun, osascript (macOS automation), xdotool/ydotool (Linux UI automation), and browser automation tools.
[CREDENTIALS_UNSAFE]: The skill handles sensitive WeChat API credentials and guides the user to store them locally.
Evidence: Prompts the user to provide WECHAT_APP_ID and WECHAT_APP_SECRET during a guided setup and stores them in plaintext .env files within the .canghe-skills/ directory.
[PRIVILEGE_ESCALATION]: The skill requires significant system permissions to automate the WeChat web interface.
Evidence: Requests the user to grant Accessibility permissions to the terminal application on macOS to enable automated keystrokes for article publishing.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-controlled markdown and HTML files, creating an attack surface for indirect injection.
Ingestion points: Markdown and HTML files provided as input to wechat-api.ts, wechat-article.ts, and wechat-browser.ts in the article publishing workflow.
Boundary markers: None identified; content is processed directly to generate WeChat drafts.
Capability inventory: Spawns subprocesses for rendering, performs network requests to download remote images specified in user content, and executes browser automation via Chrome DevTools Protocol (CDP).
Sanitization: Content is parsed using marked, but the skill performs automated actions like downloading and UI manipulation based on the content structure without exhaustive sanitization.

canghe-post-to-wechat