The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill allows the agent to ingest arbitrary content from web URLs, which represents a surface for indirect prompt injection attacks.
Ingestion points: The page_reader function, used throughout SKILL.md and in scripts/web-reader.ts, fetches raw HTML and text from external sources.
Boundary markers: The instructions do not define delimiters or provide warnings to the agent to prevent it from following instructions embedded within the extracted content.
Capability inventory: The skill includes command execution via the z-ai CLI (including file writing) and network access through the z-ai-web-dev-sdk.
Sanitization: While code examples show basic HTML tag removal, there is no logic to sanitize or filter the natural language content for malicious prompt injection patterns.
[COMMAND_EXECUTION]: The skill documentation provides various examples of shell commands using the z-ai CLI to perform web scraping, test URLs, and save output to the local file system.
[EXTERNAL_DOWNLOADS]: The skill requires the z-ai-web-dev-sdk Node.js package. It also references express and node-cron in integration examples provided in the documentation.

web-reader