The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and analyze ingestion warnings which are generated from external event data. This data is considered untrusted and could be manipulated by attackers to include malicious instructions.
Ingestion points: Data is ingested via the posthog:execute-sql tool from the system.ingestion_warnings table, specifically the details JSON field as described in SKILL.md.
Boundary markers: The skill provides clear instructions in SKILL.md under the heading 'A trust boundary that governs how you read the raw data itself', which explicitly tells the agent that details are untrusted and should not be followed as instructions.
Capability inventory: Across its various reference files, the skill utilizes impactful tools including posthog:execute-sql, posthog:project-settings-update, posthog:persons-property-delete, and posthog:cdp-functions-logs-retrieve.
Sanitization: The skill relies on natural language instructions to enforce safety rather than automated technical sanitization or strict schema enforcement for the data returned by tools.

resolving-ingestion-warnings