The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the Arize platform, such as model inputs and outputs within traces or experiment runs, which is then used to help users design evaluation templates.
Ingestion points: Data enters the agent's context through commands like ax spans export and ax experiments export as shown in the workflow sections of SKILL.md.
Boundary markers: There are no explicit instructions for the agent to use delimiters or sanitization when handling this ingested platform data.
Capability inventory: The skill can execute shell commands via the ax CLI and generate Python code for server-side evaluation tasks.
Sanitization: The skill does not prescribe specific validation or sanitization for the ingested trace data before it is presented to the user or used for template suggestions.
[DYNAMIC_EXECUTION]: The skill facilitates the creation of 'Custom Python code evaluators' which are Python classes generated by the agent and uploaded to the Arize platform for execution.
Evidence: references/cli-reference.md provides clear templates and instructions for constructing these classes and using the ax CLI flags --code and --imports to deploy them.
[COMMAND_EXECUTION]: The skill relies on the ax command-line utility for managing profiles, integrations, and evaluation workflows.
Evidence: The documentation extensively details the use of commands such as ax projects list, ax evaluators create-template-evaluator, and ax tasks trigger-run.

arize-evaluator