The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external, untrusted content including SwiftUI source code, performance trace exports, and screenshots provided by the user. This creates a surface for indirect prompt injection where malicious instructions embedded in the analyzed code or data could influence the agent's behavior.
Ingestion points: The Workflow Decision Tree and Code-First Review sections in SKILL.md explicitly request the agent to collect "Target view/feature code" and "Trace export" from the user for analysis.
Boundary markers: The instructions lack specific delimiters or "ignore" directives for the ingested content to prevent the agent from accidentally executing instructions contained within the data.
Capability inventory: The skill leverages the agent's general reasoning and code analysis capabilities to process these inputs.
Sanitization: No sanitization, validation, or filtering of the user-provided code or trace data is specified in the instructions.
[METADATA_POISONING]: The reference document 'references/wwdc-session-sources.md' contains a factually incorrect link to a non-existent future event ("WWDC 2025"), which may be misleading to users relying on the documentation for current resources.
[NO_CODE]: The skill consists entirely of Markdown instructions and reference materials, with no executable scripts, binaries, or platform-specific tool definitions included.

swiftui-performance