The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from a PostHog instance via tools such as posthog:execute-sql, posthog:query-trends, and posthog:insight-get (SKILL.md), which can contain malicious instructions within event properties or SQL results.
Ingestion points: External data is fetched and processed in the agent context during the classification and investigation steps (Step 1 and Step 3).
Boundary markers: There are no specific instructions or delimiters provided to ensure the agent ignores instructions embedded in the metric data.
Capability inventory: The skill can execute local Python scripts via subprocess calls, perform SQL operations, and write annotations or insights back to the PostHog platform.
Sanitization: No explicit sanitization or schema validation is performed on the ingested data before it is used to generate findings or as input for scripts.
[COMMAND_EXECUTION]: The skill executes local Python scripts (scripts/compare_to_prior_periods.py and scripts/breakdown_attribution.py) using python3 to analyze tool outputs. This is a functional requirement for the skill but involves a command execution surface using data-derived inputs.

investigate-metric