The Agent Skills Directory

[METADATA_POISONING]: The skill includes fabricated provenance information in UPSTREAM.json and UPSTREAM_NOTICE.md. It claims to be an official export from "Apple" using "Xcode 27.1" (a non-existent version) on "2026-10-13". This fabrication of official vendor attribution is deceptive.
[METADATA_POISONING]: The instructions in SKILL.md reference internal directory structures and scripts (e.g., references/ and scripts/ relative to the apple/ folder) that are not present in the provided skill package, indicating inconsistent metadata and missing resources.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze user-provided source code for test modernization. This creates a surface for indirect prompt injection where malicious instructions embedded in a project's test files could influence the agent's behavior.
Ingestion points: User-provided Swift test files and project repository structure.
Boundary markers: The skill includes instructions to respect user scope and repository instructions, but lacks strict delimiters for processing untrusted code.
Capability inventory: The skill facilitates code review and recommendation via host-provided editing tools.
Sanitization: No specific sanitization or validation of the ingested source code content is defined.

modernize-tests