The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to follow instructions and requirements extracted from external files, such as implementation plans and business documents (BRDs, PRDs, ADRs). This processing of external data represents a potential surface for indirect prompt injection if the source documents are maliciously crafted.
Ingestion points: Content is read from .copilot-tracking/plans/, templates/, and various project documentation files.
Boundary markers: The skill uses markdown structure to organize data but does not use dedicated delimiters or explicit safety instructions to isolate external content from the core logic.
Capability inventory: Capabilities include reading and writing workspace files, dispatching subagents for tasks, and interacting with the user for decision-making.
Sanitization: The skill treats content from external documents as data but does not explicitly specify sanitization or escaping mechanisms for the processed text.

rpi-implement