Skip to Content
Agents

Install the Skill

Agent skills are reusable instruction packages for specialized coding-agent workflows. The Quantiles agent skill follows OpenAI Codex’s agent-skill pattern  and gives coding agents a compact operating guide for using the qt CLI reliably. It standardizes how agents run evaluations, inspect and analyze results, compare eval runs, and resume interrupted workflows. It supports Codex, Claude Code, Cursor, GitHub Copilot, Gemini CLI, OpenCode, and other agents that use reusable skills or instruction files.

Prerequisites

Before installing the skill, make sure:

  • Your coding agent can read repository files and run shell commands.
  • You know where your coding agent expects local skills, reusable instruction files, or agent guidance files.
  • The agent is running from the project root or the directory where Quantiles configuration and workflow files are located.
  • If you plan to run custom Python evals, the repository has the required runtime dependencies installed.
  • If you plan to use an external model provider to evaluate (e.g., OpenAI, Anthropic, etc.), the required API credentials are configured using a configuration file.

See the troubleshooting guide for common issues and fixes.

Install the skill

Use the prompt below to set up your coding agent with the Quantiles CLI and agent skill:

Please install the Quantiles skill at github.com/quantiles-evals/skill

Alternatively, copy SKILL.md  into your agent’s skill directory.

Agent install paths may vary. If your agent uses a different skills directory, put SKILL.md wherever that agent expects reusable skill or instruction files.

Verify the install

The easiest way to validate the setup is to run a test evaluation with the built-in demo model. This runs locally, does not use your model or provider API keys, and does not incur inference costs.

Prompt your coding agent to do the following:

Run a PubMedQA benchmark and summarize the results.

After validating the installation, you can start running evaluations with your own configurations. Use the agent prompt templates as starting points for common workflows.

Last updated on