The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple sources including external URLs, PDF/Word documents, and images using LLM vision capabilities.
Ingestion points: Defined in SKILL.md and README.md as URLs, PDF/Word files, images, and folder contents.
Boundary markers: references/prompt_templates.md utilizes 【输入内容】 and {content} as delimiters; however, there are no specific instructions to the agent to disregard instructions embedded within that untrusted data.
Capability inventory: The skill possesses file system read/write access (limited to specific task and bug directories) and network capability via the gh (GitHub CLI) tool.
Sanitization: No explicit sanitization, validation, or escaping of external content is specified before interpolation into prompts.
[DATA_EXFILTRATION]: SKILL.md instructs the agent to upload bug reports to an external repository (huangwb8/bensz-bugs) using the gh tool. While this is a vendor-owned resource and the skill explicitly forbids sending credentials or private information, it represents an automated data transfer of local logs to a remote server.
[COMMAND_EXECUTION]: The skill invokes external command-line tools, specifically gh for bug reporting and bensz-collect-bugs for local logging, to perform its maintenance and reporting workflows.

research-topic-extractor