env-setup

name	env-setup
description	Use to set up a prompt-operated TDD benchmark environment before evaluation. Trigger when the user wants to choose a stack, ensure test, branch coverage, line coverage, and mutation-testing tools are available, create the workspace conventions, or write the benchmark problem statement into a prompt file for later coder-agent evaluation.

Env Setup

Purpose

Prepare the project for a TDD skill evaluation without requiring the user to edit files or run scripts manually.

This skill runs before skill-eval.

Workflow

1. Identify known setup details from the user's prompt and the repository.
2. Before asking questions, tell the user which details will be collected: stack, benchmark problem, acceptance criteria, test command, branch coverage command or proxy, line coverage command or proxy, mutation command or fallback, and constraints.
3. Use the ask-question interaction flow to collect any missing details. Do not guess high-impact benchmark intent.
4. Inspect existing build, test, coverage, and mutation tooling.
5. Establish deterministic commands for:
   • unit/integration tests
   • branch coverage or closest native equivalent
   • line/statement coverage
   • mutation testing or mutation-style proxy
6. If tools are missing, install or configure the nearest standard tool for the chosen stack when allowed by the environment.
7. Write the exact coder task into a root-level file named prompt.
8. Record setup metadata so skill-eval can run the same task across isolated worktrees.

Setup output is disposable during experimental tuning. It must be reproducible from the user's answers, but it is not expected to be preserved between iterations.

Tooling Rules

Prefer existing project conventions over introducing new tools.

If a language-specific metric is unavailable:

• Branch coverage: use condition/path coverage, decision coverage, or expected branch-case enumeration.
• Line coverage: use statement coverage or changed-file execution proxy.
• Mutation testing: use configured mutation tool if present; otherwise define deterministic mutation-style faults for evaluator review.

Always record fallbacks in setup metadata and the final report.

Prompt File Contract

Create or update a root-level prompt file with:

Problem: {problem statement}

Stack: {chosen stack}

Constraints:
- Use strict test-first TDD.
- Work in small RED/GREEN/REFACTOR cycles.
- Make minimal production changes per cycle.
- Keep verification evidence.
- Do not rely on self-report for ordering evidence.

Acceptance criteria:
{observable criteria}

Verification commands:
- test: {command}
- branch coverage: {command or proxy}
- line coverage: {command or proxy}
- mutation/test strength: {command or proxy}

Metric fallbacks:
{fallback notes}

Final Response

Report:

• selected stack
• installed or detected tools
• verification commands
• metric fallbacks
• location of the prompt file