The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill utilizes tools to monitor and ingest diagnostic events from the local SDK environment, which represents a theoretical surface for indirect prompt injection if the app being debugged generates malicious event data.
Ingestion points: MCP tools such as wait_for_events, get_sdk_state, and get_companion_boundary_diagnosis read state and event data from the application context (SKILL.md).
Boundary markers: Absent.
Capability inventory: The tools described are primarily for read-only diagnosis, status checking, and tracing.
Sanitization: The export_diagnostic_bundle functionality specifically notes the use of redaction for diagnostic data intended for support handoff.
[SAFE]: The external links point to official Meta developer documentation and a verified GitHub repository belonging to the author organization.

debugging