| name | high-evals |
| description | Manage and run coding-agent evaluations with the High-Evals CLI. Use when working in this repository to maintain prompts in prompts.json (list, add, edit, remove), choose models, and execute selected prompts in parallel or sequential mode. |
High-Evals
Run and maintain coding-agent evaluations for this repository.
Run Evaluations
- Run
high-evals run.
- Select one or more prompts from
prompts.json.
- Enter a model string.
- Choose
parallel or sequential.
- Review terminal output and generated artifacts under
evals/.
Manage Prompts
- Run
high-evals list to view prompts with indices.
- Run
high-evals add to append a prompt.
- Run
high-evals edit to modify an existing prompt.
- Run
high-evals remove to delete a prompt.
- Run
high-evals help to show command usage.
Model Input
- Use
provider/model to specify provider explicitly, for example anthropic/claude-3.5-sonnet.
- Use a plain model name to default to the
openrouter provider, for example gemini-2.5-pro.
- Run
high-evals models to query available providers and models from opencode.
- Run
high-evals models save <provider/model> to validate and save model IDs for reuse.
- Run
high-evals models save to interactively select models to save.
- Run
high-evals models saved to list saved model IDs from saved-models.json.
Files and Output
- Keep prompts in
prompts.json as a JSON array of strings.
- Keep reusable model IDs in
saved-models.json as a JSON array of strings.
- Expect each evaluation to create a timestamped folder in
evals/.
- Inspect
prompt.txt and generated files in each run folder.
Completion and Failures
- Treat a run as complete when
session.idle is received or when inactivity reaches 60 seconds.
- Check the final summary for success and failure counts.
- Use non-zero exit codes to detect failures in automation scripts.
Prerequisites
- Install Go 1.21 or later.
- Install
opencode and ensure it is available on PATH.
- Build the CLI with
go build -o high-evals . when needed.