The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides detailed instructions for querying the Cairn observability database and reading container logs using kubectl exec and kubectl logs. These data sources ingest untrusted content from external sources (e.g., HTTP request headers, application output) which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: references/observability.md (Cairn database queries via psql and kubectl logs).
Boundary markers: The skill suggests using quoted heredocs (<<'SQL') to prevent local shell expansion of queries, but does not specify delimiters or sanitization for the resulting output from logs or database rows.
Capability inventory: The agent has access to kubectl (full cluster access), ssh (host access), and git operations.
Sanitization: No specific filtering or sanitization protocols are defined for the content of processed logs before the agent interprets them.

cumulus-infra