The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses npx zai-cli throughout its instructions, which triggers the download and execution of the zai-cli package from the npm registry at runtime.
[DYNAMIC_EXECUTION]: The skill features subcommands code run and code eval (documented in references/advanced.md) which allow for the dynamic execution of arbitrary TypeScript logic.
[COMMAND_EXECUTION]: The skill relies on shell command execution to perform all its core functions, including vision analysis and repository tree exploration.
[INDIRECT_PROMPT_INJECTION]: The read command creates a vulnerability to indirect prompt injection by fetching and converting arbitrary web pages to markdown for agent processing. 1. Ingestion points: External URLs processed by the npx zai-cli read command in SKILL.md. 2. Boundary markers: No boundary markers or 'ignore' instructions are present to protect the agent from embedded malicious content. 3. Capability inventory: The skill possesses shell command execution capabilities, web search, and dynamic code evaluation. 4. Sanitization: There is no evidence of sanitization or filtering of the fetched markdown data.

zai-cli