The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes data from external system tables that may contain content influenced by other users or processes.
Ingestion points: The skill reads summary fields from system.scout_runs and metadata from document_embeddings via posthog:execute-sql (found in references/queries-yield.md and references/queries-source-parts.md).
Boundary markers: No specific boundary markers (like XML tags or delimiters) are used to isolate the data retrieved from queries when it is presented back to the agent.
Capability inventory: The skill possesses powerful write capabilities, including posthog:inbox-source-configs-partial-update, posthog:scout-config-update, and posthog:inbox-steering-instructions-update, which can modify the operational behavior of the project.
Sanitization: No explicit sanitization or instruction-filtering is applied to the retrieved data before the agent uses it to formulate recommendations.
[COMMAND_EXECUTION]: The skill makes extensive use of posthog:execute-sql to perform deep analysis of spend data. While this is central to its purpose, the ability to execute arbitrary SQL on system tables represents a high-privilege capability.
[DYNAMIC_EXECUTION]: The skill requires the agent to generate new HogQL queries at runtime for scout pre-checks and test them using posthog:scout-config-precheck-test (described in references/recommendations.md). This involves the dynamic assembly and execution of code within the database environment.

managing-self-driving-spend