The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because its core function involves processing and analyzing external telemetry, logs, and incident postmortems.
Ingestion points: The skill ingests untrusted data from multiple sources, including structured JSON logs, distributed traces, and incident learning reports (referenced in reference/logging-design.md and reference/incident-learning-postmortem.md).
Boundary markers: The skill instructions emphasize the use of OpenTelemetry semantic conventions and structured logging to provide clear data boundaries, though it does not explicitly define 'ignore' markers for embedded instructions.
Capability inventory: The skill produces implementation specifications and diagnostic command templates, including kubectl operations for log retrieval and pod management (see reference/alerting-strategy.md).
Sanitization: The skill documents robust sanitization practices, including PII scrubbing at the SDK layer and source-side log sampling to prevent data leakage (documented in reference/logging-design.md).

beacon