The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's research workflow involves ingesting and analyzing data from external, untrusted sources, which creates a potential surface for indirect prompt injection attacks.
Ingestion points: As specified in SKILL.md and rules/pre-development.md, the agent is instructed to conduct literature searches on academic platforms (Google Scholar, CNKI, IEEE, ACM, arXiv) and download/analyze PDFs, as well as investigate existing projects on GitHub.
Boundary markers: The skill references centralized safety rules (_shared/core/safety-rules.md) for agent behavior, but it does not define explicit delimiters or boundary markers for external content within the resulting planning and documentation outputs.
Capability inventory: The skill includes broad capabilities for generating executable Python code (utilizing argparse, PySide6, and Flask), creating module specifications, and generating bilingual documentation.
Sanitization: There are no instructions for sanitizing or filtering content extracted from external research materials before it is used to inform the software architecture or generated code templates.

software-planner