The Agent Skills Directory

[PROMPT_INJECTION]: The skill analyzes local codebase content and project metadata to generate task lists and proposals. This presents a theoretical surface for indirect prompt injection if files within the repository contain malicious instructions designed to influence the agent. However, the skill follows standard reasoning patterns for development tools.
Ingestion points: Reads configuration from openspec/project.md and project files via Glob, Grep, and Read tools.
Boundary markers: No specific delimiters are used to wrap external content, though the model is instructed to follow a specific analysis-driven workflow.
Capability inventory: The skill can write to the local filesystem and use WebSearch/WebFetch for external research.
Sanitization: The skill does not explicitly sanitize file content before processing it for proposal generation.
[DATA_EXFILTRATION]: The skill possesses both filesystem read capabilities and network access tools (WebSearch, WebFetch). This combination allows for a data exposure path; however, the instructions specifically scope the use of web tools to framework documentation and pattern research, which aligns with the primary purpose of the skill.

openspec-plan