The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute git log in the local repository to identify code changes during the metric anomaly window, as specified in SKILL.md step 2.3.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from PostHog query results (such as event names, property keys, and property values) to generate diagnostic reports and investigations.
Ingestion points: Data enters the agent context through tools such as posthog:query-trends, posthog:execute-sql, and posthog:insight-get as described in SKILL.md and various playbook files in the references/ directory.
Boundary markers: The skill defines a specific markdown template for findings in SKILL.md to structure the agent's output.
Capability inventory: The agent can execute HogQL queries (posthog:execute-sql), create new insights and annotations (posthog:insight-create, posthog:annotation-create), and run local shell commands (git log, python3 scripts/...).
Sanitization: The included helper scripts (scripts/compare_to_prior_periods.py and scripts/breakdown_attribution.py) parse incoming data using json.loads, but natural language content within the results is not sanitized for instructions before being processed by the LLM.

investigate-metric