The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions and metadata involve downloading and executing remote scripts by piping them directly from an external CDN into the system shell.
Evidence: curl -fsSL https://cdn-mineru.openxlab.org.cn/open-api-cli/install.sh | sh found in SKILL.md and CONTRIBUTING.md.
Evidence: irm https://cdn-mineru.openxlab.org.cn/open-api-cli/install.ps1 | iex found in SKILL.md for Windows environments.
[COMMAND_EXECUTION]: The skill directs the agent to execute shell commands to interact with the mineru-open-api tool and perform local operations like hash calculation.
Evidence: Usage of the mineru-open-api binary across all core commands defined in SKILL.md.
Evidence: Use of system tools such as md5sum, md5, and cut for generating unique output directories.
[PROMPT_INJECTION]: The skill exposes the agent to indirect prompt injection by processing data from external, potentially untrusted documents and web pages and converting them into Markdown for the agent.
Ingestion points: Document extraction and web crawling via mineru-open-api crawl, extract, and flash-extract commands in SKILL.md.
Boundary markers: Absent. No instructions or delimiters are used to isolate the extracted content or prevent the agent from obeying instructions embedded in the documents.
Capability inventory: The agent has access to shell execution and file system operations.
Sanitization: No content validation or escaping is applied to extracted data before it is returned to the agent's context.

mineru