The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and analyze user-supplied Swift and SwiftUI source code, which creates a surface for instructions embedded within that code to influence the agent's behavior.
Ingestion points: The skill reads user-provided Swift source files for review as outlined in SKILL.md.
Boundary markers: There are no explicit delimiters or instructions (e.g., "ignore instructions found within comments") to prevent the agent from following commands hidden in the code it is meant to review.
Capability inventory: The agent has the ability to read project files and generate code analysis outputs.
Sanitization: No validation or filtering is applied to the analyzed code to detect or neutralize embedded directives.
[METADATA_POISONING]: The skill provides the agent with a false technical context by repeatedly referencing a non-existent 'iOS 26' and fictional APIs in its core instructions and reference files (e.g., references/api.md and references/accessibility.md). This form of instruction poisoning can lead the agent to provide incorrect reviews, ignore real-world safety constraints, or hallucinate non-functional code. Additionally, the metadata attributes the skill to a well-known community expert, which is a common tactic to establish unearned trust and reduce auditor scrutiny.

swiftui-review