The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill executes Python code at runtime to perform Kolmogorov-Smirnov statistical tests. This is performed via inline heredoc scripts and by fetching an internal helper script, scripts/ks2.py, using the llma-skill-file-get tool. This execution is scoped to statistical calculations required for the skill's primary anomaly-detection purpose.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from PostHog event logs and insight properties, which is then used to generate user-facing reports and notebooks.
Ingestion points: Data is pulled from the event database using the execute-sql, insight-query, and insight-get tools.
Boundary markers: The instructions do not define specific lexical delimiters to separate untrusted event data from the agent's instructions during report generation.
Capability inventory: The skill can execute SQL queries, run Python code, and author reports/notebooks via the scout-emit-report and notebooks-create-markdown tools.
Sanitization: The skill implements statistical guards (such as minimum absolute count floors and seasonality matching) but does not specifically sanitize string content for potential prompt injection vectors.
[COMMAND_EXECUTION]: The skill utilizes Bash to coordinate the execution of the Python interpreter for running its statistical analysis logic.
[PERSISTENCE]: The skill maintains a durable "watchlist" and baseline history using the platform's scratchpad tools (scout-scratchpad-remember, scout-scratchpad-search). This allows the agent to maintain state and track metric performance across separate execution sessions.

signals-scout-anomaly-detection