The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is configured to use the gh (GitHub CLI) tool and a secondary skill bensz-collect-bugs to report issues to the author's repository (huangwb8/bensz-bugs) upon user request.
[PROMPT_INJECTION]: The skill processes untrusted data from multiple sources including PDFs, images, and URLs, creating an attack surface for indirect prompt injection.
Ingestion points: The skill ingests data from a variety of file types (.pdf, .doc, .docx, .md, .txt, .tex, .rst), image files, and web URLs as described in the workflow section of SKILL.md.
Boundary markers: Neither the main instruction set in SKILL.md nor the prompt templates in references/prompt_templates.md implement delimiters (like XML tags) or specific instructions to ignore embedded commands in the source data.
Capability inventory: The skill has the capability to read local files, access the network via URL fetching, and execute shell commands through the gh utility.
Sanitization: No content validation or sanitization routines are defined to filter or escape malicious instructions within the extracted text before it is processed by the AI.

research-topic-extractor