The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill utilizes external search tools to gather documentation and research patterns, creating a potential surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context via mcp__perplexity__search and mcp__context7__query-docs as specified in the MCP Research section of SKILL.md.
Boundary markers: There are no explicit instructions or delimiters provided to the agent to treat search results as untrusted content or to ignore embedded instructions within those results.
Capability inventory: The skill allows Write and Edit operations (defined in SKILL.md), which could be influenced by malicious instructions discovered during external searches.
Sanitization: No sanitization or validation logic is defined for processing the output of the search tools before it is used to guide file modifications.
Unverifiable Dependencies (SAFE): The skill references standard, well-known packages such as Grpc.AspNetCore and Google.Protobuf. The use of the buf CLI for linting and code generation is standard industry practice for Protocol Buffers and does not pose an atypical risk in this context.

protobuf-design