The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a workspace setup script (scripts/setup_workspace.py) and instructions for the agent to execute various Python and Node.js utilities for tracing, sampling, and running application evaluations locally.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process application evidence, such as logs and traces, which may contain content from untrusted users.
Ingestion points: Ingests traces, support examples, and datasets from local files or managed API exports.
Boundary markers: The instructions explicitly advise redacting content at collection boundaries and warn that escaping prevent execution but does not anonymize data.
Capability inventory: Includes capabilities for file writing, network calls to vendor APIs, and local command execution across Python and Node.js environments.
Sanitization: The skill provides guidance on redacting sensitive information and using side-effect-free fixtures for evaluation runs.
[DYNAMIC_EXECUTION]: The setup_workspace.py script performs runtime assembly of project scripts by reading template files and substituting hardcoded placeholder paths with the user's workspace root before writing the final scripts to the filesystem.
[EXTERNAL_DOWNLOADS]: The skill references documentation, SDKs, and CI tools hosted on the vendor's domain (confident-ai.com) and GitHub organization (github.com/confident-ai).

eval-build