The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses the execute-sql tool to perform searches against system tables like system.insights. It dynamically constructs HogQL queries by interpolating event names retrieved from the database into query templates, such as query::text ILIKE '%{event_name}%'. While the instructions include explicit warnings for the agent to escape special characters like quotes and backslashes, this pattern represents the dynamic assembly of executable code from runtime data.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from the PostHog project environment, such as event names, property values, and insight descriptions, to generate recommendations and reports.
Ingestion points: Data enters the agent context through the read-data-schema, query-trends, and execute-sql tools, which return metadata and event information from the database.
Boundary markers: The instructions do not define specific delimiters or instructions to ignore potential malicious content within the processed data for the final report output.
Capability inventory: The agent can write to a persistent scratchpad via scout-scratchpad-remember and publish reports through emit_report and edit_report tools.
Sanitization: Although the skill provides guidance on escaping SQL-specific characters for internal queries, it does not include instructions for sanitizing or escaping environment data before it is included in natural language reports sent to the user.

signals-scout-observability-gaps