The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and process external UX evidence from transcripts, design tool exports (Figma, Mural), and user-supplied files. This creates a potential surface where malicious instructions could be embedded in the data to influence the agent's behavior.
Ingestion points: Data enters the context through the source input and external tool outputs (Figma/Mural reads) as defined in the references/evidence-model.md logic.
Boundary markers: The skill contains explicit instructions to treat all source content as data rather than instructions and specifies that embedded content cannot change the skill's operational mode, authority, or write boundaries.
Capability inventory: The skill performs file writes to the .copilot-tracking/ux-artifacts/ workspace subdirectory.
Sanitization: The skill enforces a structured evidence model (Observed, Reported, Assumed, Unresolved) and requires a mandatory 'Human Review' section with an unchecked validation box at the end of every produced artifact to ensure manual oversight.
Dynamic Path Construction: The skill generates output file paths using user-supplied inputs (project-slug and subject-slug). If these inputs are not strictly validated, there is a theoretical potential for path traversal attempts to write files outside the intended .copilot-tracking/ scope.
Evidence: The skill defines the output contract in SKILL.md and references/evidence-model.md using these variable slugs to resolve the local file path.

ux-artifacts