The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple sources, including external server error messages (latest_error), user-authored SQL queries (query_log), and external webhook metadata. While the skill includes explicit instructions for the agent to treat this data as non-authoritative, the combination of processing raw external input and possessing capabilities like SQL execution and report emission creates a standard vulnerability surface for indirect prompt injection.
Ingestion points: Data is ingested in SKILL.md from the system.source_schemas.latest_error field, the query_log.query field (containing raw user SQL), and the external-data-sources-webhook-info-retrieve tool.
Boundary markers: The skill contains a dedicated section titled 'Untrusted data — errors, table names, and source labels' which provides instructions to treat these strings as data, not instructions, and to prioritize trusted identifiers (UUIDs) over names.
Capability inventory: The skill has access to execute-sql for querying warehouse tables and the scout-emit-report and scout-edit-report tools for creating or modifying user-visible reports.
Sanitization: The skill instructs the agent to quote external content as short snippets and to truncate long messages to avoid payload echoes.

signals-scout-data-warehouse