The Agent Skills Directory

[METADATA_POISONING]: The skill uses deceptive versioning, referencing 'iOS 27', 'Xcode 27', and 'WWDC 2026', which do not exist. Furthermore, all documentation links are pointed to an unofficial third-party domain ('sosumi.ai') while presenting them as authoritative Apple documentation. This could lead users to rely on unverified or potentially malicious information sources.
[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow for ingesting external telemetry data that is not fully sanitized. Ingestion points: The skill instructions and code in SKILL.md and references/metrickit-patterns.md consume data from MetricManager.metricReports and MetricManager.diagnosticReports async sequences. Boundary markers: No specific delimiters or safety instructions are used to separate the external telemetry data from the agent's control flow. Capability inventory: The skill provides a blueprint for durable storage and network upload of these reports, involving potential file system writes and network transmissions. Sanitization: The data is processed using JSONEncoder/JSONDecoder, but there is no semantic validation of the content to prevent instructions embedded in report fields from influencing the agent's behavior.
[DATA_EXFILTRATION]: The skill provides instructions for building a 'Durable Export and Upload' pipeline. While intended for performance telemetry, this pattern describes the collection and transmission of detailed application state, call stacks, and environment metadata to external servers, which could be repurposed for unauthorized data exfiltration.

metrickit